This reminds me of a (completely different, but felt equally) bizzare problem with ghostty (the terminal emulator) on Linux and Steam, specifically Baldur's Gate 3.
At the start of the game after an update, BG3 compiles the (Vulkan) shaders. To avoid hammering the computer, it checks the load, including I/O pressure, and then throttles itself. It turns out that Ghostty uses io_uring, and even if it's not doing anything this somehow counts as pressure. This was made worse by the fact that I usually have a dozen or so Ghostty windows.
End result, BG3 compiles shaders using 1 core, throttled, instead of all available 20 cores. Instead of a minute or so, it'd take several hours.
(For closure: I configured ghostty to use epoll instead. Probably not optimal but works around the problem.)
If I am reading the thread correctly, the io_uring maintainer's position is that applications trying to use the metric that "poll via io_uring" handles differently from "poll via epoll or select" are fundamentally misguided and measuring something that doesn't make sense, so they don't want to slow down io_uring to placate those apps.
gavinsyancey 3 minutes ago [-]
> iowait is fundamentally incompatible with sane accounting in the first
place, that is the problem. It really tells you nothing, and was
deprecated since the days of uniprocessor systems. The problem here is
not how iowait is done in the kernel, it's that people are using this
metric in the first place. Which is why you just want to NOT do that,
and do sane accounting on your disk drives, if you care about that kind
of thing.
> Which is why I refuse to cater to any kind of iowait accounting metrics,
it's just adding overhead to support a metric that is largely
nonsensical in the first place. It doesn't answer any questions that an
application/tier owner would want to know. Those questions might be
things like "am I I/O bound?" or "is the device I'm using being slow?".
I wish people would realize that, rather than continue to place value in
iowait as a metric for ANYTHING, and stop trying to work around it.
It sounds more like their position is "iowait is a useless metric" and that polling with io_uring completely breaks iowait metrics but fixing it would add a minor slowdown so they refuse to do so because they don't think it's a valuable metric to begin with.
dustbunny 3 hours ago [-]
Amazing. How did you figure this out?!
simonask 4 hours ago [-]
I hate this. We have a whole operating system doing its absolute best to prioritize hardware resources. Don't be out here throttling yourself.
Especially for something like compiling shaders. It literally never happens unless the user is about to go play a game, which is going to consume most resources anyway. To the user, there's nothing more important running right now.
kevin_thibedeau 1 hours ago [-]
My motherboard crashes when I peg all the cores. Presumably it can't maintain the core voltage under load. Vendors need to guard against these scenarios by being extra cautious because it's a hard problem to diagnose that they'll get blamed for.
Dylan16807 53 minutes ago [-]
Guarding against that seems like a bad idea, but even then just use a couple less threads than physical cores. Don't get too clever.
You in particular should probably disable a couple cores until your hardware stops being defective. Have you looked into updates and voltage droop compensation settings?
0c3ca83 38 minutes ago [-]
That sounds like a faulty motherboard. Return it.
Arainach 18 minutes ago [-]
Users don't care that it's a problem with the motherboard. They have a working computer, they buy a game and it doesn't work, they blame and return the game.
Users can be unreasonable, but in this case I agree - if I reported an issue to a vendor and their response was "return your motherboard" I wouldn't work with that vendor again.
BenjiWiebe 15 minutes ago [-]
Sounds like they don't have a working computer. Computer users occasionally do things that end up pegging all cores, so I'm pretty sure that someone in that case would know their computer has some kind of issue.
MichaelZuo 39 minutes ago [-]
It’s just reality that motherboards can malfunction or degrade in peculiar ways that a manufacturer’s quick boot up checks cant always catch. No matter the reliability claims.
If they want to cater to customers who foolishly reject reality and take every claim from manufacturer’s at face value… then foolishly misaimed blaming seems expected?
But if they just roll over and swallow the foolish blaming without even challenging customers to double check, then they are also facilitating more nonsense.
cowlevel 3 hours ago [-]
Especially on Linux, the default throttling just doesn't work very well.
embedding-shape 4 hours ago [-]
> To the user, there's nothing more important running right now.
Well, to a certain degree. I have a relatively uncommon GPU, and run games on Steam (and Linux), so literally every game does shader compilation for me. Even with a beefy CPU and beefy GPU, if the shader compilation uses more than 16 threads, my computer basically freezes until its done with it, as it's so performance intensive, kind of have to throttle things if I want to be able to use the desktop for the 10 minutes the shader compilation takes sometimes.
Edit: "Do you use Linux? That only happened to me when I was using a dram-less ssd for swap and doing IO heavy operations on the NFS mounted drive"
I do! I do have NFS drives, but not for Steam games or anything else relevant. I'm also not using swap at all. RAM seems fine during this, it's just pegging all cores to 100% usage immediately and stays there, making everything else suffer.
Aurornis 3 hours ago [-]
> Even with a beefy CPU and beefy GPU, if the shader compilation uses more than 16 threads, my computer basically freezes until its done with it
Strange. This never happens to me on Windows. I only notice that the fans spin up a little because all cores are pegged.
jack_pp 3 hours ago [-]
i've only seen the shader compilation on linux.
braiamp 4 hours ago [-]
Do you use Linux? That only happened to me when I was using a dram-less ssd for swap and doing IO heavy operations on the NFS mounted drive, since the system will use ram as a buffer (which pushes things to swap)
Hendrikto 4 hours ago [-]
> Even with a beefy CPU and beefy GPU, if the shader compilation uses more than 16 threads, my computer basically freezes
Sounds like a scheduling problem that should be adressed at the OS level.
embedding-shape 4 hours ago [-]
Seems a deep one if so. I've tried a bunch of alternative schedulers to try to avoid the issue, like Bore and BMQ, but seems to happen regardless. Been like this on previous computers too, fwiw, unless I turn down the concurrency for shader compilation stuff will lag heavily until it finishes.
bilegeek 3 hours ago [-]
> tried a bunch of alternative schedulers
1. Have you tried enabling full preemption as well? I run Debian, and before I did that I had lots of stutters from any kind of background load thanks to dynamic preemption.
2. Even then, ZFS was causing stutters until I lowered it's priority as well (spl_taskq_thread_priority=0)
3. On non-ZFS drives, try setting your I/O scheduler to Kyber for NVME or BFQ for everything else. "mq-deadline" isn't great for interactive use on a SATA SSD, even for non-boot drives, and even NVME sometimes stutters under "none" for some people.
> I'm also not using swap at all.
4. zswap (combined with zstd compression, "min_ttl_ms=1000", and "vm.page-cluster=0") and "defer+madvise" memory compaction can help. (Something about the kernel really liking having swap on-hand.)
Steam runs shader compilations at batch policy and nice 19, so without RAM or I/O bottlnecks even gaming shouldn't be too bad (though from personal experience it still induces some stuttering). It's just most distros are tuned horribly for this. (Not to sound flippant, I understand the frustration of "just try this" and it doesn't work, but it's everything I can think of RN.)
jordanb 2 hours ago [-]
You could try cpu pinning or ulimits to limit the program from using one core so there's always one free.
But it's also possible that the problem is in the kernel and the shaders are actually doing some system call that's locking up the kernel. Like maybe there's a bug in your cuda or nvidia driver that's causing the kernel to lock.
sayotte 50 minutes ago [-]
io_uring you say… one of the most infamous subsystems for security, best reserved for stringent performance requirements. Why on earth would Ghostty default to that?
BenjiWiebe 12 minutes ago [-]
Maybe because one of ghostty's selling points is performance.
Also if you're running untrusted code that attacks your not-very-common terminal emulator, you're probably in more danger from the "untrusted code" part. Threat model and all that stuff.
manarth 3 hours ago [-]
Better description: "My USB network switch was cutting my download speed".
The fact that it's a monitor is a little bit irrelevant.
xp84 2 hours ago [-]
Same old adapter worked perfectly if the monitor was taken out of the path. Kind of an interaction thing with both, really.
dcrazy 49 minutes ago [-]
The twist is it would happen with any additional device in the signal path. The mere presence of a hub causes the USB control message to take slightly longer to reach the adapter—about the same amount of time it takes to receive 5 Ethernet frames at 2Gbps.
The fix is to cut the overhead by sending more than 4 frames per interrupt. Unfortunately that required buying a new adapter with a different chip.
alt227 15 minutes ago [-]
The fix is actually to plug it directly into the host, and not use a hub. Very common thing to need to do with cheap or older USB Hardware. Paying more money for something new when the old one works fine is pointless and wasteful.
intrasight 1 hours ago [-]
> 2026 Studio Display XDR, with everything plugged into it so switching laptops is one cable
I think I'd have accepted having two cables from the get-go. And if I observed this problem, the first diagnostic I would do is to plug the Ethernet directly into the laptop.
Fewer technical pieces in a puzzle sometimes trumps ease of use.
pavinjoseph 59 minutes ago [-]
I view anything "USB" with more than a little suspicion after my hellish journey into running RAID over USB drives.
amelius 2 hours ago [-]
We're talking about an Apple user here. They think in terms of appliances, and "everything just works".
jdiff 60 minutes ago [-]
> So it’s the display, right? Well, I would find out: sort of yes, sort of no
Your assertion doesn't seem to align with the article itself.
disjxiwjciwjd 16 minutes ago [-]
Please tell me you didn’t just go the insanely childish “Apple users are dumb” route.
throw0101d 3 hours ago [-]
Meta:
> Somehow I am one of those few very lucky people to have the opportunity of owning a 2.5 gig symmetrical line for my internet.
I'm not going to complain against more bandwidth (as a GenXer I stared my online journey at (IIRC) 9600bps (maybe lower): much celebration with V.32/V.90), but I'm curious to know how many residential use-cases can take advantage of that much bandwidth. Linux ISOs?
How many services allow that much speed on a single stream / (TCP) session? Maybe you could have five at 500 Mbps?
Certainly downloading security updates or games may be quicker, but I would think that above 500Mbps (and certainly 1 Gbps) there would be major diminishing returns. Is anyone regularly getting close to that utilization in their daily work / hobbies (and not just a momentary spike on patch Tuesday)?
Zyst 3 hours ago [-]
Torrents, and things like downloading games from Steam feel really good if you have a very strong connection. A lot of games can be downloaded in the time it takes to go grab a drink from the fridge. It adds some nice quality of life, there isn't as much need to keep a large downloaded set - just download when you need it, and it'll be there quickly.
I think going from something like 100 Mb download to 1 Gb is slightly overrated. You hit heavy diminishing returns. Symmetric Upload, for me, is what really changes your quality of life a lot. Uploading a youtube video suddenly isn't an event, it takes a few seconds. Want to upload an image in any chat? Done before you consciously see it uploading. Heavy commits are instant, even with a bunch of resources. It changed my behavior. I started uploading more stuff to remote servers because the cost in time got so much lower.
Guillaume86 3 hours ago [-]
Yup as someone on a 1000/20 link (can’t get anything better in my little village in BE), decent upload is far more useful than more download after a point (far before 1gbps).
Aurornis 3 hours ago [-]
The major CDNs routinely deliver full throughput.
It’s great when experimenting with LLMs because I can quickly try new models and download from the internet almost as fast as if coming from my NAS.
Game updates are also a great use case. I don’t game often so when my friends want to play I often have to update the game. Some patches are 60GB or more. When they come down at full line speed that’s valuable time I’m getting back to spend with my friends in game.
bloopernova 11 minutes ago [-]
Downloading big models is much nicer on a 1Gbit fiber line. Uploading test docker images too if you're forced into modifying them locally.
coryrc 23 minutes ago [-]
Some games are hundreds of GB of disk space and also have a terrible patching algorithm where they need to make an entire duplicate of all files to patch them, despite downloading less than a GB of patch. So you have to uninstall stuff. With fast Internet you can actually use most of your drive because you can uninstall then download a new install which only needs X space not 2X.
nunez 2 hours ago [-]
2.5Gbit (or 5Gbit in my case) is INFINITELY helpful if you run k8s clusters. Container images download as quickly as moving a big file on a spinning rust hard drive even if someone else is watching TV or doing equally intensive stuff.
kakacik 1 hours ago [-]
Watching TV is nothing, even 4K netflix stream is around 15-20 mbit and shouldn't spawn many concurrent threads unlike say bittorrent.
xp84 2 hours ago [-]
Do not underestimate the possibilities. I have 1.0/0.1Gb cable, and doing remote access is a complete joke. Even streaming a 1080p movie up from my house is extremely challenged.
This also locks me out of hosting my own “cloud” storage at home where it belongs - I don’t want to be out and waiting forever to get a file, so I stick to traditional cloud services for now. Slower than local storage on the LAN of course though.
A fiber isp is digging trenches soon. Cannot wait!
boston_clone 2 hours ago [-]
I mean, yeah, you have a 100 Mbit/s (or 12.5 MB/s) upload rate under ideal conditions. If my math is right - haven’t had my coffee yet - that’s just 4% of the OPs bandwidth.
nojs 47 minutes ago [-]
I also have this on paper. It’s surprising how much home hardware caps out at 1 Gbps or below. I had to replace several routers and run cables everywhere to get the full speed. Once you add on tailscale, a VPN, etc it’s a proper challenge to get it consistently working above 1Gbps.
WhiteDawn 3 hours ago [-]
99% of the time I don’t need it. However, when I do need it, it sure is nice to move quickly during updates or downloading sdk/assets.
Even better is those can saturate as much bandwidth as possible while the rest of the family are uninterrupted with Netflix or video calls.
makeitdouble 3 hours ago [-]
> Maybe you could have five at 500 Mbps?
Let's say you're on a Meet call with a dozen of people, one of them presenting their screen to explain an ongoing incident. You're following along while checking your dashboards and code on a remote server. You're of course keeping Slack sessions in the background.
I don't think you'd be saturating a 2.5G bandwidth, but perhaps about half of it ?
If your family is also using it in parallel that could get close to full use.
Aurornis 3 hours ago [-]
> Let's say you're on a Meet call with a dozen of people, one of them presenting their screen to explain an ongoing incident. You're following along while checking your dashboards and code on a remote server. You're of course keeping Slack sessions in the background.
This is closer to 1-10Mbps than 2.5Gbps. Slack is negligible bandwidth. Video conferencing is high compression, low bandwidth video.
The 1G and 2.5G lines benefit bulk file downloads.
jayd16 2 hours ago [-]
Maybe so but even at 10Mbps that's a lot of the 20Mbps upload you get from a residential cable line.
eska 1 hours ago [-]
A high quality 60fps 1080p gameplay stream is 4Mbps. That mostly still frame powerpoint presentation is completely negligible, and the 6 thumbnail sized videos you see at 30fps are too. You can run a whole office of people on that call on smaller connections.
rayiner 3 hours ago [-]
That wouldn’t come close to saturating 100 mbps. Screen sharing even at 4k is ridiculously efficient, maybe 4-5 mbps for what you’re describing where most of the image isn’t changing most of the time.
throw0101d 3 hours ago [-]
> Let's say you're on a Meet call with a dozen of people […]
I have been on Zoom and Google Meets calls with that many people on my laptop which has a Wifi connection of ≤400 Mbps (sometimes <200Mbps). You don't need >500 Mbps for online meetings.
Are you in meetings where everyone is sending 4Kp60?
silvestrov 2 hours ago [-]
I (for one) would definitely NOT want to see the faces of my colleagues in 4K.
lowres, please.
only 4K if the feed is from a full room with +10 people in it.
penagwin 3 hours ago [-]
I think steam lets you get to at least 2gbps.
An interesting side effect is that bufferbloat becomes less of a problem - you’re more likely able to pull 1gbps without spiking your ping.
But to be honest unless you have a household with multiple people, it’s something you mainly notice occasionally with updates, backups, large assets, etc.
NBJack 3 hours ago [-]
I doubled my measly 20 MPBps upload speed to 40 simply because of how long it was taking me to get rendered assets to someone. Having higher bandwidth than that (eventually, I hope) would be a real help in video editing.
rayiner 3 hours ago [-]
I have 10 gig into my house, and most services can’t saturate it. Microsoft is crap, big game downloads don’t even top out the gigabit link in the xbox. Apple gets 2-3 gig downloading xcode updates and the like.
nunez 2 hours ago [-]
I wish I could get 10Gbit; friggin' AT&T lol.
Fetching from Google or anything on S3 can definitely fill your pipe
seiferteric 3 hours ago [-]
My cable is I think up to 2Gbps now and I regularly see game updates in steam hit like 1.4gbps. Probably it is closely located in a CDN I would assume.
zeristor 3 hours ago [-]
I do homebrew updates regularly, and quite often it can take more than an instant. On my 40Mb connection it’s slow.
Run it in background, don’t update so often perhaps; I guess I just update as tic.
IshKebab 60 minutes ago [-]
Yeah basically nothing. I have 900 Mbps and pretty much the only things where the other end gets about 250 are Steam, Docker images, and Bittorrent. You really don't need more than 250 or at most 500 Mbps for the vast vast majority of cases.
monster_truck 3 hours ago [-]
You definitely notice it
vachina 2 hours ago [-]
4K HDR BRrip in 1 minute
fancyfredbot 4 hours ago [-]
Since lots of comments misunderstood the article, the extra latency from going through the monitor meant his laptop couldn't acknowledge the packets quickly enough. That led a buffer on the ethernet adapter to fill and that led it to drop packets.
A different adapter which required less frequent acknowledgement fixed the problem for him.
It is quite a nice post which gets to the bottom of incredibly opaque behaviour of USB hubs.
jayd16 1 hours ago [-]
The fix is that it could send more packets per ack not that it needed to happen less. If the buffer was bigger but still capped at 4 packets per ack, the full buffer drain speed would remain at ~2.
alansaber 4 hours ago [-]
Probably a stupid question, but why is USB behaviour so opaque?
jayd16 2 hours ago [-]
There's not much monitoring built into usb hubs. It's actually far less when you think about all the monitoring a network switch has.
frizkie 36 minutes ago [-]
I had the same problem recently when trying to figure out high bandwidth game streaming from my desktop to my Steam deck in the living room. The beta streaming tech is “pyrowave” and it is super low latency and looks INCREDIBLE compared to traditional codecs, but as a trade off it is extremely sensitive to packet loss, and uses an appropriately large amount of bandwidth (7-800mbit at 4k120). Wired network isn’t _strictly_ a requirement but it might as well be, and the steam deck doesn’t have a physical port - but the steam deck dock does.
Anyway, the exact same situation arises, where some network condition or physical configuration (there’s a lot of hardware in between these two devices, including MoCA adapters) meant that the network adapter buffer fills up and starts dropping packets. The buffer size is configurable through some terminal command and maxing that out makes the issue disappear immediately.
gh123man 3 hours ago [-]
I have nothing useful to add other than I feel so validated reading this. Ive been struggling with this issue on and off for over a year. With the release of Astra, it was able to finally unstick me to the point of enabling flow control which has been good enough.
I experienced this problem with a macbook and a thunderbolt dock. I had tried multiple different brands of docks and ethernet adapters, and various macbooks. One interesting thing I discovered is if I plugged a non-thunderbolt cable between the dock and the laptop, I would achieve full speed with no packet loss, so interestingly the issue doesn't occur over pure USB but only over thunderbolt.
This has got to be a bug on Apple's side right? Would be nice if they fixed it.
dcrazy 36 minutes ago [-]
What kind of Thunderbolt cable, and what kind of USB cable? The downside of using the USB-C connector for everything is that it obscures the difference between active and passive Thunderbolt cables.
xp84 2 hours ago [-]
Oof. That is the kind of thing where Apple will never do anything. They implement standards, but they guarantee absolutely nothing if even so much as one cable involved isn’t Apple-branded. Using docks and Ethernet adapters? Good luck! They’ll just ask if you’ve considered buying a $5000 monitor with a $1000 stand to use as a dock instead. (And ironically, that’s what OP did and still had the problem :D)
glitchc 16 minutes ago [-]
I suspect the problem here is the UGreen ethernet adapter. I have soured on UGreen after a couple of their recent hardware bits proved flaky and one damaged two laptop motherboards (it was a usb-c hub).
zodwick 2 hours ago [-]
>It’s run by this one guy
This was super surprising, have always thought ISPs are a super hard businesses. But good for you mate, good ISPs are lifesavers.
I have had the opposite experience with my ISP guys. One of the VMs I got as part of work only had IPv6 and to my shock my internet didn't have one. When I contacted my ISP they legit told "You can either have IPv4 or IPv6, but not both". Unfortunately I was on a 12 month plan and couldn't easily cancel so ended up working around using tailscale.
eska 1 hours ago [-]
There are communities of small towns and villages running their own ISP and providing help materials for others who want to try.
zodwick 40 minutes ago [-]
Do you have any pointers ? Idt this is a thing where I live tbh
throw0101d 3 hours ago [-]
What's the difference in BOM cost for Realtek RTL8153 (1GigE) versus RTL8156 (2.5G) versus RTL8157 (5G) versus RTL8159 (10G)?
Palomides 2 hours ago [-]
chips are probably like $2 to $50 across that range, hard to tell since realtek sucks
srcreigh 6 hours ago [-]
Nice post. Unsatisfying conclusion - did you ever figure out why the original adapter was throttled when plugged into the monitor, and not when plugged into the computer?
cibyr 5 hours ago [-]
The additional latency from going through the monitor made the bandwidth-delay product exceed the amount of data the adapter was able to send before receiving an ack. I thought that was fairly clear from the explanation, but now I notice that he didn't mention the latency without the monitor.
nh2 3 hours ago [-]
That doesn't seem to be described by the article at all.
The latency added in the monitor is described as around 8 us, completely negligible compared to any Internet connection's latency.
So the latency without the monitor should be completely irrelevant.
db48x 4 minutes ago [-]
Negligible compared to the RTT, sure. But they’re talking about the latency between when the network adaptor signals that a packet has arrived and when the CPU actually services the interrupt to tell the card to send the packet(s) to main memory. The NIC’s buffer could only hold 4 packets, so by the time the NIC got a DMA buffer to copy the packets into a fifth or sixth packet had arrived and been dropped.
nh2 2 hours ago [-]
The article answers this
"if the speed of the drain from my adapter to my Mac is half a gig slower than the speed of the input, then you can imagine the buffer getting filled more and more until it overflows. Those are dropped packets, which result in a hard backoff when the router resends them, so you drop loads in speed. And that’s the reason for all of this."
The monitor decides to send the adapter's packets on at only 2 Gbit/s, which leads to buffer overflow, thus packet loss, thus e.g. TCP congestion-control backoff.
The only thing we don't know is _why_ the monitor chooses to limit the bandwidth to those 2.07 Gbit/s (computed as the 41 KiB blocks per (7.6+8.2) µs). If it sent e.g. 81 KiB blocks, doubling the bandwidth, the problem wouldn't happen.
That'd be between Apple and the monitor.
archerx 6 hours ago [-]
I’m guessing it’s a USB bandwidth issue. I learned about these kind of limitations when I tried to have 3 kinects running at the same time on the same USB bus.
swiftcoder 5 hours ago [-]
> I’m guessing it’s a USB bandwidth issue
It's explicitly not a bandwidth issue (that would also affect the SSD). It's a latency issue - the chip doesn't have a big enough buffer to keep sufficient packets in flight at once.
nh2 3 hours ago [-]
> It's explicitly not a bandwidth issue
That's explicitly the opposite of what the author says.
"the speed of packets coming into the adapter is 2.5 Gbps. That’s what the router pushes in. The speed going out is around 2 Gbps."
Of course latency is always on the denominator side in the "bandwidth = data / time" equation, but it seems wrong to claim it's latency when it could also be improved by fixing the numerator; it's the entire fraction that's insufficient, thus bandwidth. As written:
"The reason is that if you add up the average time it takes to get the go-ahead through my monitor’s hub, which is around 7.6 µs, to the average time it takes to send 4 KB, around 8.2 µs, you can find what the maximum speed is"
And later, increasing the numerator is exactly what fixes it with the other hardware: "I can see that instead of sending 4 1 KB packets per burst, it would send 8!"
> the chip doesn't have a big enough buffer to keep sufficient packets in flight at once
How does this make sense?
Even if his buffer was 100x larger, if the fill bandwidth exceeds the drain bandwidth, it'll eventually fill up.
The article says that explicitly:
"Well, if the speed of the drain from my adapter to my Mac is half a gig slower than the speed of the input, then you can imagine the buffer getting filled more and more until it overflows."
swiftcoder 2 hours ago [-]
> Even if his buffer was 100x larger
"buffer" was perhaps a poor choice of words on my part, because "buffer" is ambiguous. In this case I am referring to the notional "transmit buffer", which is 4 in the non-working case and raised to 8 in the working adapter.
> "I can see that instead of sending 4 1 KB packets per burst, it would send 8!"
Increasing the number of packets in flight does not change the bandwidth, it only improves latency sensitivity (i.e. you can fire more data before requiring an ACK).
> it seems wrong to claim it's latency when it could also be improved by fixing the numerator.
Increasing the bandwidth without increasing the transmit buffer would not fix this issue. The adapter he ends up buying does both (allow 2x the packets in flight, and 2x the bandwidth).
The bottom line here is that the original adapter handles full duplex 2.5g when plugged into the laptop directly, and the SSD gets full duplex bandwidth in excess of 2.5g through the same intermediate link (i.e. the display), so both the adapter and the display have sufficient bandwidth. The added hop of latency however makes the short transmit buffer insufficient to cover the latency.
chrisweekly 2 hours ago [-]
This primer on latency and bandwidth is 13 years old but still worth sharing (for the sake of all readers)
He states that high speed SSDs through the monitor are fine though.
prepend 5 hours ago [-]
Uploads were fine.
linker3000 1 hours ago [-]
Used to get this with dial-up modems when they were either placed on top of the CRT (just to get very toasty?), or the incoming phone line ran too close to some CRTs.
This would reduce the established connection speed due to electrical noise from the monitor.
cozzyd 3 hours ago [-]
Would bbr congestion control help here to.get speeds to 2 Gbps?
nh2 2 hours ago [-]
Seems likely. BBR (v1) does not do the hard 30% "multiplicative decrease" on packet loss (caused by the frame drop in the article's diagram), so I'd expect it to be able to correctly estimate the bandwidth to the ~2 Gbit/s bottlenecked by the monitor-Mac connection.
cozzyd 1 hours ago [-]
Yeah, I'd expect it too, but doing some research it seems Apple still doesn't support BBR. But I wonder if a speedtest with quic would show better results for the author.
iamflimflam1 3 hours ago [-]
I recently tracked down some poor network performance to some dodgy Ethernet cables in my patch panel.
Sure we are talking about a different level in the networking stack and very different protocols. But with every component you add things get less synchronous. Well, no buffering needs to involved by the middleman, the mechanism for the throttling could also be different. What do I know...
The 35 KiB buffer from the article is not very large, and the problem would exist unmodified if it was smaller.
As the article explains, "if the speed of the drain from my adapter to my Mac is half a gig slower than the speed of the input, then you can imagine the buffer getting filled more and more until it overflows", and this would hold for any buffer, smaller or larger.
Bufferbloat causes primarily latency problems, not bandwidth problems.
Bandwidth problem would occur at most as secondary effects, e.g. if a packet is delayed by a buffer for so long that the sender-expected timeout for an ACK expires, it considers the packet lost, and throttles down sending. But that's not happening here.
Pxtl 3 hours ago [-]
What's your handle on Sauerbraten?
tymscar 24 minutes ago [-]
Tymscar, as everywhere else! I don’t play often enough
jmyeet 4 hours ago [-]
A couple of thoughts:
1. These 1 cable solutions are nice but (IME) incredibly finnicky. I've gone through several Caldigit docks that have died;
2. When you think about it, these docks (and monitors) are really 100+ Gb/s switches. It's kind of no wonder that there are odd behaviors;
3. IME that cable that provides power to the laptop and handles power/video (you know, so it really is 1 cable) is the absolute worst. Whenever you buy them, buy 2-3 because I can almost guarantee 1 will be faulty and it'll be weird, annoying problems eg the screen gling black for a split second randomly;
4. I did a lot of testing with 1GbE USB-Ethernet adapters and it's true that they are not all created equal. In a perfect world, they should do ~930Mb/s. Depending on which one you get it'll be as low as 880 (or lower); and
5. I would honestly expect better from Apple. With the price premium you pay for these Studio monitors, I would expect the hub part to be pretty much seamless.
ferdzo 5 hours ago [-]
I've encountered a similar problem, the wifi on a MacBook Air was dying once connected to a Dell monitor over USB-C. The wifi signal was going from full to nearly no signal. Connecting to the same monitor with a dongle and HDMI didn't have the same outcome. Never had a time to find the culprit. Probably EMI or bad grounding.
gdevenyi 3 hours ago [-]
USB3 has huge EMI issues that haven't been very discussed. It's messing with your signal.
entrope 5 hours ago [-]
What a click-bait headline. The conclusion was that it was due to having a more bottlenecked USB-to-Ethernet adapter path. Would the same result have held with a discrete Thunderbolt 5 hub, either with a built-in 2.5 Gbps or faster Ethernet port or with the original USB-to-Ethernet adapter? Would it have held with another USB hub? Who knows -- just blame it on the expensive monitor!
rustyhancock 5 hours ago [-]
In a sense it felt like the obvious outcome of a "one cable" setup.
It's unreasonable assumption to expect such a setup to manage multiple full speed connects in my opinion.
If I was running support it and later heard on I just unplugged the ethernet from my monitor direct to my laptop and the speed improved I'd dispair.
ricardobeat 4 hours ago [-]
> It's unreasonable assumption to expect such a setup to manage multiple full speed connects
The bandwidth in question is nowhere near saturating even a single USB3 connection.
jayd16 2 hours ago [-]
And yet...
It's a lot easier to hit real world limits than theoretical spec limits.
dan_grec 4 hours ago [-]
But now it is doing exactly that.
ricardobeat 4 hours ago [-]
The explanation is not that clear, but you missed the point. It is a networking issue known as 'bandwidth-delay product': the combination of latency and bandwidth is what limits the speed, and in this case the extra latency from the monitor hop was the culprit. Not the adapter. The faster adapter is a workaround.
entrope 18 minutes ago [-]
You sound like Steve Jobs: "the adapter is fine, you just plugged it in wrong." No: if the buffers are too small to tolerate an intervening USB hub, the peripheral was designed poorly.
Edited to add: Unlike networks in general, the USB standard imposes maximum diameters and latencies, so designers know what to design for. If the higher-spec adapter also had awful goodput, then it might be reasonable to point fingers at the monitor.
field10DOTcom 5 hours ago [-]
[dead]
eska 1 hours ago [-]
My whole wifi slowed to 5kb/s because of a bluetooth smartwatch interfering with the 2.5ghz wifi. I was always told bluetooth and wifi sharing a spectrum could not possibly ever be problematic. Needless to say I hate Bluetooth even more since then.
Sopel 5 hours ago [-]
> So I thought I would do what any ~~human~~ apple user would do, and I spent more money on a new adapter.
Moto7451 4 hours ago [-]
Most people probably would buy another adapter. Most people would probably not debug the USB connection for its latency and throughput over USB.
alt227 3 hours ago [-]
Most Apple users would do this. Most other people would have just used wifi on the laptop and an HDMI cable for plugging their monitor in.
Sopel 3 hours ago [-]
or have RJ45 on hardware they want to connect to the internet via cable
cozzyd 3 hours ago [-]
I do wish my think pad's Ethernet port was 2.5G...
Betelbuddy 56 minutes ago [-]
Unnecessarily long...TL;DR: The monitor introduced a tiny delay in USB communication, which caused the Ethernet adapter to lose data packets...
luxuryballs 3 hours ago [-]
I'm always jealous of people who have the willpower and time to investigate this kind of stuff, I would have plugged it into the laptop and moved on! It's perhaps the most nerdy "grass must be greener" luxury I can imagine (in a good way).
Jingling9188 2 hours ago [-]
What is this clickbait headline? It's the monitor hub and/or the USB adapter they were using.
taneq 4 hours ago [-]
Hah, well if my monitor at home ever does this to me I can take my revenge by slowly removing my jacket, or sitting down on my chair. (Samsung Odyssey G9, it's ultra-sensitive to ESD anywhere near it so if I'm wearing the right jumper I can move my chair or wave my arms around and the capacitative heresy will make it wig out for a good 2-3 seconds.)
At work we run HP Z40c monitors and while they've been bulletproof in terms of image quality, on my old laptop the mouse cursor was definitely more laggy when running in a one-cable setup. We're using Logitech MX Master combos so I just moved the dongle from the monitor to my laptop and that fixed the lag issue.
Wow this sounds like a shill post. If anyone from HP or Logitech wants to talk message me. Otherwise Logitech Opti+ or whatever its called can fk off.
candl 4 hours ago [-]
I hate this attitude:
"I wasn’t getting three quarters of the speed I was paying for"
No, you were. It was not the ISPs fault.
I get constant support requests like that. People install their own equipment, adapters, wifi repeaters, sketchy unmanaged switches and keep complaining about speeds when the ONT/CPE whatever works perfectly and the connection is fine.
All it takes is to connect to the CPE directly and do a measurement.
It's like people calling the power plant operator directly if one of their light bulbs is dimmer than the rest.
alt227 3 hours ago [-]
> I hate this attitude
Well yours seems pretty great!
TacticalCoder 4 hours ago [-]
I've only read the headline: USB degrading speeds?
(five minutes passed, I skimmed through TFA)
Ah yup, of course it's somehow related to USB.
It's like the infamous line: "It's always DNS" (and when it's not it's the firewall).
When download speeds are degrading like that, it's always some weird USB thing.
In the earlier days people didn't know that, for example, plugging a monitor into an USB hub could, as a result, make the rest degrade to USB 2.0 speeds. Unplug the monitor, HUB is speedy again now that it doesn't need to deal with that 4K / 60 Hz or whatever monitor data.
It's not a criticism or anything and I'm not saying that's what's going on here: I'm just saying USB has many gotchas and there can be, as in TFA and as in many other (document or not) cases, weird things happening.
At times (and doubly so when troubleshooting), you're better to simply try to plug your USB cable directly in the computer instead of in a USB hub / monitor USB hub. (EDIT: ah yup, ofc eventually he try that)
There are thing too complex to ever be secure and then there are things too complex to not have gotchas.
USB is... Wild?
P.S: and then of course in addition to this USB wildness there's the mad insanity of USB... Cables.
learntoplaypitc 2 hours ago [-]
[flagged]
field10DOTcom 5 hours ago [-]
[dead]
enoch_hou 41 minutes ago [-]
[dead]
nonacosa 3 hours ago [-]
[flagged]
nonacosa 4 hours ago [-]
[flagged]
53 minutes ago [-]
thefz 4 hours ago [-]
I knew it was Apple even before clicking
skibble 3 hours ago [-]
No, it was his Ethernet adapter.
alt227 3 hours ago [-]
When he plugged the original ethernet adapter straight into the laptop it worked as expected. It was the monitor not supporting that specific setup and chip.
aaronmdjones 2 hours ago [-]
It has nothing to do with the monitor's support. The GP is correct; the fault is in the Ethernet adapter. Specifically, it doesn't have a buffer large enough to sustain 2.5 Gbps when going via the monitor because the latency is too high. You'd have the same problem on many USB hubs and it would still be the Ethernet adapter's fault. It could have used a larger buffer or it could send more data at a time to the host so that the buffer doesn't fill up, but it doesn't.
alt227 40 minutes ago [-]
I would say if a USB adapter works fine in a laptop USB port, then there is nothing wrong with it. If you are trying to use it in a monitor and it introduces so much latency that doesnt work, then thats a different issue. Its not a faulty adapter.
dcrazy 33 minutes ago [-]
The adapter manufacturer sourced a part that could not sustain the product’s advertised bandwidth when connected through a USB hub. USB hubs are not esoteric. This is the fault of the adapter manufacturer cheaping out on a part.
tymscar 16 minutes ago [-]
I feel like I am going a bit crazy reading some of these comments. I don’t know why people don’t read the whole blog post to understand that the monitor works great and it gets maximum speeds when doing things like transferring files from my ssd.
The adapter works great and I am getting maximum speeds when connected to my laptop.
The laptop works great and I get maximum speeds when doing any of those things.
The hub inside the monitor works great and I get maximum speeds through it even with multiple things plugged in at the same time.
None of the parts are broken. Together they don’t work as they are supposed to because of some super intricate process of how usb works when it comes to acknowledging packages. That’s why I wrote the post. The title is true and yeah, it’s sensational but that’s what titles are there for. It’s not a false clickbait.
dcrazy 11 minutes ago [-]
I read the entire post and enjoyed it very much. Thank you for writing it.
That said, I think you’re being too charitable to UGREEN. They shouldn’t advertise a product as 2.5Gb/s if it can’t actually transfer 2.5Gb/s to the host when the signal path includes a hub.
tymscar 8 minutes ago [-]
You are probably right. I think the reason why I cut them the slack is because of how cheap the adapter is and that it does give me the speed when connected directly or if I go with flow control. Its a weird one
alt227 29 minutes ago [-]
I think we are all arguing the same thing right? The adapter worked fine when used directly in a device but not when latency was high due to other devices in the chain?
I have had hundreds of USB devices in the past which dont work on a USB hub, its very common. The first troubleshooting step in basically any USB devices manual is if you are having trouble and you are using a hub, remove the hub and plug directly into the device.
The user went way over the top in troubleshooting this issue. Thats the first thing I would have tried. Its not worth the post and the title should really be something like "USB device didnt work with a hub, but did directly in the host" and it would never have made it to the front page of HN other than maybe to be mocked.
dcrazy 26 minutes ago [-]
No, you are blaming the monitor manufacturer for the customer not getting the advertised speed out of their Ethernet adapter when they connected it through the monitor’s integrated USB hub.
I am saying your blame is misplaced, and that the adapter manufacturer deserves the blame because they designed their product around a controller whose USB implementation has so much inherent latency that it mathematically cannot sustain the product’s advertised throughput when connected through a hub. The manufacturer should have tested this extremely common scenario and picked a part that could sustain it.
alt227 9 minutes ago [-]
No, Im blaming the user for overengineering and sensationalising the fact that a cheap USB device didnt work on a hub. That is a problem as old as USB itself and to go this far down the rabbit hole on it is laughable. There are plenty of other people in the thread that agree.
> The manufacturer should have tested this extremely common scenario and picked a part that could sustain it.
Hard disagree. Maybe if you bought a better one you would expect that, but I would expect a $15 2.5Gb ethernet adapter to not work on a hub right off the bat tbh.
Rendered at 17:52:55 GMT+0000 (Coordinated Universal Time) with Vercel.
At the start of the game after an update, BG3 compiles the (Vulkan) shaders. To avoid hammering the computer, it checks the load, including I/O pressure, and then throttles itself. It turns out that Ghostty uses io_uring, and even if it's not doing anything this somehow counts as pressure. This was made worse by the fact that I usually have a dozen or so Ghostty windows.
End result, BG3 compiles shaders using 1 core, throttled, instead of all available 20 cores. Instead of a minute or so, it'd take several hours.
(For closure: I configured ghostty to use epoll instead. Probably not optimal but works around the problem.)
If I am reading the thread correctly, the io_uring maintainer's position is that applications trying to use the metric that "poll via io_uring" handles differently from "poll via epoll or select" are fundamentally misguided and measuring something that doesn't make sense, so they don't want to slow down io_uring to placate those apps.
> Which is why I refuse to cater to any kind of iowait accounting metrics, it's just adding overhead to support a metric that is largely nonsensical in the first place. It doesn't answer any questions that an application/tier owner would want to know. Those questions might be things like "am I I/O bound?" or "is the device I'm using being slow?". I wish people would realize that, rather than continue to place value in iowait as a metric for ANYTHING, and stop trying to work around it.
It sounds more like their position is "iowait is a useless metric" and that polling with io_uring completely breaks iowait metrics but fixing it would add a minor slowdown so they refuse to do so because they don't think it's a valuable metric to begin with.
Especially for something like compiling shaders. It literally never happens unless the user is about to go play a game, which is going to consume most resources anyway. To the user, there's nothing more important running right now.
You in particular should probably disable a couple cores until your hardware stops being defective. Have you looked into updates and voltage droop compensation settings?
Users can be unreasonable, but in this case I agree - if I reported an issue to a vendor and their response was "return your motherboard" I wouldn't work with that vendor again.
If they want to cater to customers who foolishly reject reality and take every claim from manufacturer’s at face value… then foolishly misaimed blaming seems expected?
But if they just roll over and swallow the foolish blaming without even challenging customers to double check, then they are also facilitating more nonsense.
Well, to a certain degree. I have a relatively uncommon GPU, and run games on Steam (and Linux), so literally every game does shader compilation for me. Even with a beefy CPU and beefy GPU, if the shader compilation uses more than 16 threads, my computer basically freezes until its done with it, as it's so performance intensive, kind of have to throttle things if I want to be able to use the desktop for the 10 minutes the shader compilation takes sometimes.
Edit: "Do you use Linux? That only happened to me when I was using a dram-less ssd for swap and doing IO heavy operations on the NFS mounted drive"
I do! I do have NFS drives, but not for Steam games or anything else relevant. I'm also not using swap at all. RAM seems fine during this, it's just pegging all cores to 100% usage immediately and stays there, making everything else suffer.
Strange. This never happens to me on Windows. I only notice that the fans spin up a little because all cores are pegged.
Sounds like a scheduling problem that should be adressed at the OS level.
1. Have you tried enabling full preemption as well? I run Debian, and before I did that I had lots of stutters from any kind of background load thanks to dynamic preemption.
2. Even then, ZFS was causing stutters until I lowered it's priority as well (spl_taskq_thread_priority=0)
3. On non-ZFS drives, try setting your I/O scheduler to Kyber for NVME or BFQ for everything else. "mq-deadline" isn't great for interactive use on a SATA SSD, even for non-boot drives, and even NVME sometimes stutters under "none" for some people.
> I'm also not using swap at all.
4. zswap (combined with zstd compression, "min_ttl_ms=1000", and "vm.page-cluster=0") and "defer+madvise" memory compaction can help. (Something about the kernel really liking having swap on-hand.)
Steam runs shader compilations at batch policy and nice 19, so without RAM or I/O bottlnecks even gaming shouldn't be too bad (though from personal experience it still induces some stuttering). It's just most distros are tuned horribly for this. (Not to sound flippant, I understand the frustration of "just try this" and it doesn't work, but it's everything I can think of RN.)
But it's also possible that the problem is in the kernel and the shaders are actually doing some system call that's locking up the kernel. Like maybe there's a bug in your cuda or nvidia driver that's causing the kernel to lock.
Also if you're running untrusted code that attacks your not-very-common terminal emulator, you're probably in more danger from the "untrusted code" part. Threat model and all that stuff.
The fact that it's a monitor is a little bit irrelevant.
The fix is to cut the overhead by sending more than 4 frames per interrupt. Unfortunately that required buying a new adapter with a different chip.
I think I'd have accepted having two cables from the get-go. And if I observed this problem, the first diagnostic I would do is to plug the Ethernet directly into the laptop.
Fewer technical pieces in a puzzle sometimes trumps ease of use.
Your assertion doesn't seem to align with the article itself.
> Somehow I am one of those few very lucky people to have the opportunity of owning a 2.5 gig symmetrical line for my internet.
I'm not going to complain against more bandwidth (as a GenXer I stared my online journey at (IIRC) 9600bps (maybe lower): much celebration with V.32/V.90), but I'm curious to know how many residential use-cases can take advantage of that much bandwidth. Linux ISOs?
How many services allow that much speed on a single stream / (TCP) session? Maybe you could have five at 500 Mbps?
Certainly downloading security updates or games may be quicker, but I would think that above 500Mbps (and certainly 1 Gbps) there would be major diminishing returns. Is anyone regularly getting close to that utilization in their daily work / hobbies (and not just a momentary spike on patch Tuesday)?
I think going from something like 100 Mb download to 1 Gb is slightly overrated. You hit heavy diminishing returns. Symmetric Upload, for me, is what really changes your quality of life a lot. Uploading a youtube video suddenly isn't an event, it takes a few seconds. Want to upload an image in any chat? Done before you consciously see it uploading. Heavy commits are instant, even with a bunch of resources. It changed my behavior. I started uploading more stuff to remote servers because the cost in time got so much lower.
It’s great when experimenting with LLMs because I can quickly try new models and download from the internet almost as fast as if coming from my NAS.
Game updates are also a great use case. I don’t game often so when my friends want to play I often have to update the game. Some patches are 60GB or more. When they come down at full line speed that’s valuable time I’m getting back to spend with my friends in game.
This also locks me out of hosting my own “cloud” storage at home where it belongs - I don’t want to be out and waiting forever to get a file, so I stick to traditional cloud services for now. Slower than local storage on the LAN of course though.
A fiber isp is digging trenches soon. Cannot wait!
Even better is those can saturate as much bandwidth as possible while the rest of the family are uninterrupted with Netflix or video calls.
Let's say you're on a Meet call with a dozen of people, one of them presenting their screen to explain an ongoing incident. You're following along while checking your dashboards and code on a remote server. You're of course keeping Slack sessions in the background.
I don't think you'd be saturating a 2.5G bandwidth, but perhaps about half of it ?
If your family is also using it in parallel that could get close to full use.
This is closer to 1-10Mbps than 2.5Gbps. Slack is negligible bandwidth. Video conferencing is high compression, low bandwidth video.
The 1G and 2.5G lines benefit bulk file downloads.
I have been on Zoom and Google Meets calls with that many people on my laptop which has a Wifi connection of ≤400 Mbps (sometimes <200Mbps). You don't need >500 Mbps for online meetings.
Are you in meetings where everyone is sending 4Kp60?
lowres, please.
only 4K if the feed is from a full room with +10 people in it.
An interesting side effect is that bufferbloat becomes less of a problem - you’re more likely able to pull 1gbps without spiking your ping.
But to be honest unless you have a household with multiple people, it’s something you mainly notice occasionally with updates, backups, large assets, etc.
Fetching from Google or anything on S3 can definitely fill your pipe
Run it in background, don’t update so often perhaps; I guess I just update as tic.
A different adapter which required less frequent acknowledgement fixed the problem for him.
It is quite a nice post which gets to the bottom of incredibly opaque behaviour of USB hubs.
Anyway, the exact same situation arises, where some network condition or physical configuration (there’s a lot of hardware in between these two devices, including MoCA adapters) meant that the network adapter buffer fills up and starts dropping packets. The buffer size is configurable through some terminal command and maxing that out makes the issue disappear immediately.
I experienced this problem with a macbook and a thunderbolt dock. I had tried multiple different brands of docks and ethernet adapters, and various macbooks. One interesting thing I discovered is if I plugged a non-thunderbolt cable between the dock and the laptop, I would achieve full speed with no packet loss, so interestingly the issue doesn't occur over pure USB but only over thunderbolt.
This has got to be a bug on Apple's side right? Would be nice if they fixed it.
This was super surprising, have always thought ISPs are a super hard businesses. But good for you mate, good ISPs are lifesavers.
I have had the opposite experience with my ISP guys. One of the VMs I got as part of work only had IPv6 and to my shock my internet didn't have one. When I contacted my ISP they legit told "You can either have IPv4 or IPv6, but not both". Unfortunately I was on a 12 month plan and couldn't easily cancel so ended up working around using tailscale.
The latency added in the monitor is described as around 8 us, completely negligible compared to any Internet connection's latency.
So the latency without the monitor should be completely irrelevant.
"if the speed of the drain from my adapter to my Mac is half a gig slower than the speed of the input, then you can imagine the buffer getting filled more and more until it overflows. Those are dropped packets, which result in a hard backoff when the router resends them, so you drop loads in speed. And that’s the reason for all of this."
With diagram https://blog.tymscar.com/studio-display-ethernet/bottleneck....
The monitor decides to send the adapter's packets on at only 2 Gbit/s, which leads to buffer overflow, thus packet loss, thus e.g. TCP congestion-control backoff.
The only thing we don't know is _why_ the monitor chooses to limit the bandwidth to those 2.07 Gbit/s (computed as the 41 KiB blocks per (7.6+8.2) µs). If it sent e.g. 81 KiB blocks, doubling the bandwidth, the problem wouldn't happen.
That'd be between Apple and the monitor.
It's explicitly not a bandwidth issue (that would also affect the SSD). It's a latency issue - the chip doesn't have a big enough buffer to keep sufficient packets in flight at once.
That's explicitly the opposite of what the author says.
"the speed of packets coming into the adapter is 2.5 Gbps. That’s what the router pushes in. The speed going out is around 2 Gbps."
Of course latency is always on the denominator side in the "bandwidth = data / time" equation, but it seems wrong to claim it's latency when it could also be improved by fixing the numerator; it's the entire fraction that's insufficient, thus bandwidth. As written:
"The reason is that if you add up the average time it takes to get the go-ahead through my monitor’s hub, which is around 7.6 µs, to the average time it takes to send 4 KB, around 8.2 µs, you can find what the maximum speed is"
And later, increasing the numerator is exactly what fixes it with the other hardware: "I can see that instead of sending 4 1 KB packets per burst, it would send 8!"
> the chip doesn't have a big enough buffer to keep sufficient packets in flight at once
How does this make sense? Even if his buffer was 100x larger, if the fill bandwidth exceeds the drain bandwidth, it'll eventually fill up. The article says that explicitly:
"Well, if the speed of the drain from my adapter to my Mac is half a gig slower than the speed of the input, then you can imagine the buffer getting filled more and more until it overflows."
"buffer" was perhaps a poor choice of words on my part, because "buffer" is ambiguous. In this case I am referring to the notional "transmit buffer", which is 4 in the non-working case and raised to 8 in the working adapter.
> "I can see that instead of sending 4 1 KB packets per burst, it would send 8!"
Increasing the number of packets in flight does not change the bandwidth, it only improves latency sensitivity (i.e. you can fire more data before requiring an ACK).
> it seems wrong to claim it's latency when it could also be improved by fixing the numerator.
Increasing the bandwidth without increasing the transmit buffer would not fix this issue. The adapter he ends up buying does both (allow 2x the packets in flight, and 2x the bandwidth).
The bottom line here is that the original adapter handles full duplex 2.5g when plugged into the laptop directly, and the SSD gets full duplex bandwidth in excess of 2.5g through the same intermediate link (i.e. the display), so both the adapter and the display have sufficient bandwidth. The added hop of latency however makes the short transmit buffer insufficient to cover the latency.
https://hpbn.co/primer-on-latency-and-bandwidth/
This would reduce the established connection speed due to electrical noise from the monitor.
Sure we are talking about a different level in the networking stack and very different protocols. But with every component you add things get less synchronous. Well, no buffering needs to involved by the middleman, the mechanism for the throttling could also be different. What do I know...
The 35 KiB buffer from the article is not very large, and the problem would exist unmodified if it was smaller.
As the article explains, "if the speed of the drain from my adapter to my Mac is half a gig slower than the speed of the input, then you can imagine the buffer getting filled more and more until it overflows", and this would hold for any buffer, smaller or larger.
Bufferbloat causes primarily latency problems, not bandwidth problems.
Bandwidth problem would occur at most as secondary effects, e.g. if a packet is delayed by a buffer for so long that the sender-expected timeout for an ACK expires, it considers the packet lost, and throttles down sending. But that's not happening here.
1. These 1 cable solutions are nice but (IME) incredibly finnicky. I've gone through several Caldigit docks that have died;
2. When you think about it, these docks (and monitors) are really 100+ Gb/s switches. It's kind of no wonder that there are odd behaviors;
3. IME that cable that provides power to the laptop and handles power/video (you know, so it really is 1 cable) is the absolute worst. Whenever you buy them, buy 2-3 because I can almost guarantee 1 will be faulty and it'll be weird, annoying problems eg the screen gling black for a split second randomly;
4. I did a lot of testing with 1GbE USB-Ethernet adapters and it's true that they are not all created equal. In a perfect world, they should do ~930Mb/s. Depending on which one you get it'll be as low as 880 (or lower); and
5. I would honestly expect better from Apple. With the price premium you pay for these Studio monitors, I would expect the hub part to be pretty much seamless.
It's unreasonable assumption to expect such a setup to manage multiple full speed connects in my opinion.
If I was running support it and later heard on I just unplugged the ethernet from my monitor direct to my laptop and the speed improved I'd dispair.
The bandwidth in question is nowhere near saturating even a single USB3 connection.
It's a lot easier to hit real world limits than theoretical spec limits.
Edited to add: Unlike networks in general, the USB standard imposes maximum diameters and latencies, so designers know what to design for. If the higher-spec adapter also had awful goodput, then it might be reasonable to point fingers at the monitor.
At work we run HP Z40c monitors and while they've been bulletproof in terms of image quality, on my old laptop the mouse cursor was definitely more laggy when running in a one-cable setup. We're using Logitech MX Master combos so I just moved the dongle from the monitor to my laptop and that fixed the lag issue.
Wow this sounds like a shill post. If anyone from HP or Logitech wants to talk message me. Otherwise Logitech Opti+ or whatever its called can fk off.
Well yours seems pretty great!
(five minutes passed, I skimmed through TFA)
Ah yup, of course it's somehow related to USB.
It's like the infamous line: "It's always DNS" (and when it's not it's the firewall).
When download speeds are degrading like that, it's always some weird USB thing.
In the earlier days people didn't know that, for example, plugging a monitor into an USB hub could, as a result, make the rest degrade to USB 2.0 speeds. Unplug the monitor, HUB is speedy again now that it doesn't need to deal with that 4K / 60 Hz or whatever monitor data.
It's not a criticism or anything and I'm not saying that's what's going on here: I'm just saying USB has many gotchas and there can be, as in TFA and as in many other (document or not) cases, weird things happening.
At times (and doubly so when troubleshooting), you're better to simply try to plug your USB cable directly in the computer instead of in a USB hub / monitor USB hub. (EDIT: ah yup, ofc eventually he try that)
There are thing too complex to ever be secure and then there are things too complex to not have gotchas.
USB is... Wild?
P.S: and then of course in addition to this USB wildness there's the mad insanity of USB... Cables.
None of the parts are broken. Together they don’t work as they are supposed to because of some super intricate process of how usb works when it comes to acknowledging packages. That’s why I wrote the post. The title is true and yeah, it’s sensational but that’s what titles are there for. It’s not a false clickbait.
That said, I think you’re being too charitable to UGREEN. They shouldn’t advertise a product as 2.5Gb/s if it can’t actually transfer 2.5Gb/s to the host when the signal path includes a hub.
I have had hundreds of USB devices in the past which dont work on a USB hub, its very common. The first troubleshooting step in basically any USB devices manual is if you are having trouble and you are using a hub, remove the hub and plug directly into the device.
The user went way over the top in troubleshooting this issue. Thats the first thing I would have tried. Its not worth the post and the title should really be something like "USB device didnt work with a hub, but did directly in the host" and it would never have made it to the front page of HN other than maybe to be mocked.
I am saying your blame is misplaced, and that the adapter manufacturer deserves the blame because they designed their product around a controller whose USB implementation has so much inherent latency that it mathematically cannot sustain the product’s advertised throughput when connected through a hub. The manufacturer should have tested this extremely common scenario and picked a part that could sustain it.
> The manufacturer should have tested this extremely common scenario and picked a part that could sustain it.
Hard disagree. Maybe if you bought a better one you would expect that, but I would expect a $15 2.5Gb ethernet adapter to not work on a hub right off the bat tbh.