The End of TCP for AI Clusters — John Ousterhout, Stanford

Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.

Original language: English

Original video: The End of TCP for AI Clusters — John Ousterhout, Stanford · Channel: AI Engineer · 18:26 · Watch on YouTube

Topics: Model inference and optimization, AI infrastructure and hardware

Summary

John Ousterhout argues that AI networking workloads are shifting from large transfers, where throughput dominates, toward more small-message exchanges in inference and agentic applications, where latency—especially tail latency—can limit GPU utilization and overall throughput. He explains how incast congestion, sender-based congestion control, and stream-oriented designs can delay short messages in TCP and RDMA. He introduces Homa, a message-based transport protocol that uses receiver-driven congestion control, prioritizes shorter messages, and leverages switch priority queues. In the presented benchmark, Homa’s short-message P99 latency is about 13 times faster than TCP, while its longest-message performance is also nearly twice as good.

Key points

Large transfers have historically dominated AI workloads, making throughput the main networking concern.
Inference and agentic workloads increasingly involve small messages for metadata, coordination, and synchronization.
When GPUs wait for synchronized messages, high tail latency can leave compute resources idle and reduce throughput.
Incast congestion can create queues at the destination’s last network hop, delaying short messages behind large transfers.
TCP and RDMA rely on sender-side congestion control, which Ousterhout says can react slowly and cause queue buildup and rate oscillations.
TCP and RDMA use a byte-stream model that obscures message boundaries and can cause head-of-line blocking.
Homa is message-based and treats request-response remote procedure calls as its fundamental unit.
Homa uses receiver-driven congestion control: receivers issue grants that regulate when senders transmit subsequent packets.
Homa prioritizes short messages using shortest remaining processing time and modern switches’ priority queues.
In the shown benchmark, Homa’s short-message P99 latency is about 13 times faster than TCP; its longest-message performance is nearly twice as good.

Chapters

Start	Chapter	Summary
00:01	Why AI networking needs low latency	Ousterhout introduces the changing demands of AI workloads and argues that latency is becoming increasingly important.
02:16	From large transfers to latency-sensitive messages	He contrasts throughput-oriented training transfers with small-message exchanges in inference and agentic workloads, explaining how synchronization delays can stall GPUs.
06:01	Incast congestion and the limits of TCP and RDMA	He describes how incast creates queues, why sender-based congestion control struggles to respond, and how stream semantics lead to head-of-line blocking.
11:28	How Homa manages messages and congestion	Ousterhout presents Homa’s message-based design, receiver-driven grants, short-message prioritization, and use of switch priority queues.
15:44	Benchmark results and closing remarks	He compares Homa and TCP latency results and encourages developers to test whether short-message latency is limiting their applications.

Quotes

what really matters is tail latency. — 03:51

TCP and RDMA are just not well suited to this environment. — 10:58

Homa prioritizes shorter messages. — 13:05

Homa is less than 100 microseconds, about 13 times faster. — 16:49

is high latency for short messages affecting your throughput? — 17:20

Concepts and entities

Homa, TCP, RDMA, incast, SRPT, ECN, John Ousterhout, Ben Moazeni, Stanford University

Full transcript (original language)

00:01 [music] >> Please welcome to the stage the Professor Emeritus at Stanford University, John Ousterhout.

00:37 >> Good morning. It’s really great to be here to talk about the network side of AI applications and in particular to make the case that latency matters and is probably going to be mattering more in the future. But I just want to say this is a talk is unusual for me. I’ve never before given a talk where there are fog generators in the auditorium. Just a really San Francisco experience, I guess. So it’s it’s well known that AI workloads depend on really great networking performance in order to achieve their own performance. Of course, that’s because the workloads are so large that they have to be

01:10 distributed across machines and then you have to communicate between the machines. But what I want to talk about today is it it seems that those workloads are changing. And so I hope to do three things over the next 15 or 20 minutes. First, to convince you that in fact the workloads are changing and that whereas the workloads used to be completely dominated by large transfers where throughput is the key metric that matters, that we’re seeing more and more smaller transfers where the latency is crucial. The second thing I hope to do is to convince you that legacy protocols like

01:44 TCP and RDMA are poorly suited to this environment. They weren’t designed for this environment and unfortunately they suffer from very high tail latency when you mix small messages with large ones. And talk a little bit about why that’s the case. Then third, I’d like to introduce Homa, which is a new protocol we’ve developed at Stanford that actually was designed in a clean slate redesign to handle data center workloads like these. And in fact, it does quite well on those workloads and can reduce tail latency by an order of magnitude or more. So, I’ll tell you a little bit about Homa.

02:16 So, let’s dive in. First, workloads. Historically, AI workloads have consisted of enormous transfers between machines. And that’s all that really mattered. Gigabytes of data for things like weight gradients and and so on. In these workloads, what you really care about is throughput. How many gigabits per second you can pump through the pipes. And these are relatively easy workloads for networks because if it takes a while to set up the connection and start the transfer, it doesn’t matter. The

02:48 transfers go on for so long that all that really matters is the throughput. And so, in these environments, TCP and RDMA perform pretty well. Uh by the way, when I say RDMA, what I really mean is Rocky RDMA over converged Ethernet, which is the underlying transport that’s used by RDMA for most purposes today. So, anyhow, the old workloads, big transfers, throughput matters, uh the legacy protocols work pretty well. However, it appears that the workloads are changing. They’re becoming more granular with smaller chunks of

03:21 computation and smaller exchanges of data. And this seems to be particularly true in the world of inference and also in agentic workloads. Not so much for training workloads are still massive transfers. And so, what’s happening is that more and more there are small message exchanges, typically for things like metadata and coordination, such as checking to see if a particular entry is present in a KV cache that’s distributed, or doing barrier synchronization at the end of periods of compute. And for these workloads,

03:51 what really matters is latency. That is, what’s the round trip time to send some small piece of data across the network, do a little bit of computation, and get a small result back again? And in fact, it isn’t just just latency or average latency that matters. What really matters is tail latency. That is, you’d like to know that if we send a whole lot of small messages, all of them will complete quickly. So, for example, we typically measure things like 99th percentile latency. And if we have high tail latency, that

04:22 can limit the overall throughput of the system. So, here’s an example. Suppose common thing is to take a workload and split it up across several nodes, which do intensive computation using their GPUs for some period of time. And then once they’ve all finished their computation, you do some small exchange between the nodes to exchange data and metadata, and then they’ll go on to the next round of computation. And while that exchange is happening, that synchronization is happening, the GPUs are sitting idle. So, if even one of those exchanges takes a

04:55 long time, it turns out the whole process stalls. You need all of those exchanges to complete before you can go on to the next phase of computation. Now, if the computation phase is, say, 5 seconds, and it takes a few milliseconds for the exchange, you know, not a not a problem. And that’s historically what it’s been. But now with the agentic workloads, where you’re trying to pump out tokens relatively rapidly at a regular rate, the periods of computation are getting down into sort of the millisecond time scale. And if it also takes milliseconds to do that synchronization,

05:28 then you’re wasting a significant fraction of your GPU’s resources waiting for the the synchronization to occur. So, I’m curious. I’d like to just do a quick audience poll here. Is there anybody here where you have reason to believe that the latency of small messages is impacting the overall throughput of your applications? If so, can you just raise your hand? See if there’s anybody out there today. Actually, more hands than I expected. So, quite a few people out there are raising their hands. I think this problem is likely to get worse as the trends continue.

06:01 So, what’s going on? Why is tail latency bad? Well, typically the cause is congestion resulting from incast. So, incast is when several nodes all decide simultaneously to transfer data to some destination node. And if they all send large messages, well, the links are the same everywhere in the network. So, three nodes can transfer three times as fast as one node can possibly receive. And so, what happens is that packets accumulate at the last hop going to that destination in the top of your

06:35 access switch at its egress port for the destination node. Then, if some other node decides it wants to send a short message to that same destination, the short message gets stuck behind the long ones in the queue there. And actually, as that causes delay, and in the worst case, so many packets arrive that the switch runs out of buffer space that it has to drop packets, and then there are there timeouts and retransmissions that make everything even worse. So, somehow we need some way to reduce the congestion in those queues. Somehow

07:10 we have to get the sending nodes to stop sending so fast so the queues don’t just build up without limit. So, the way this is done historically virtually all network protocols before Homa, including TCP and RDMA, congestion control is the responsibility of the sender. So, senders somehow have to figure out that congestion is happening, and then they have to slow down their rate of transmission. Now, you you might wonder, why are senders doing it? Cuz the congestion is way over at the other end of the data center network. How does the the sender find out?

07:43 Well, in the in the old really old days, the way they would find out is the queues would overflow and packets would get dropped. The sender would detect the packets got lost cuz it wouldn’t get acknowledgements back, and it would assume that means there’s congestion and then slow down its rate of transfer. That’s really expensive. So, today there are better techniques that mostly involve the switches providing information. So, a top of rack switch, when it sees that the queue length for an egress port has reached some threshold or starting to fill long before the queue overflows, it starts marking all of the packets to

08:15 pass through with what’s called early congestion notification, ECN marking. And so, when those packets pass through to the receiver, the receiver sees the marking in the packets, and then when it communicates back to the sender next, for example, to send an acknowledgement, then it includes that marking that goes back to the sender. And now the sender sees the sender realize, “Oh, there’s congestion someplace. I’ve got to slow down my rate of transmission.” So, that’s the basic idea. Unfortunately, getting this right is really hard, really hard. It’s very hard

08:47 for the congestion to figure out exactly how to set its rates cuz it gets one bit of information. There’s congestion someplace. And there are multiple senders all sending to the same destination. They’re all trying to make adjustments simultaneously. How much do you cut back? And how do I know when I can ramp up again? And even worse, it’s really hard to do this in a way that’s stable because there’s control lag. That is, it takes time before the sender finds out that there’s congestion. And in fact, using this process, it typically takes several round trips for the sender to gradually adjust its rate

09:20 to get just the right rate to match the available bandwidth. But by the time you do that, in a network that things have changed. New transmissions have started or old ones have finished. And so, these systems tend to never stabilize. They’re constantly oscillating between sending too too and and too little. And this problem’s been around for a long time. It’s been known in the research community for more than 20 years now. There’ve been tons of papers published on it. There have been some improvements made, that’s undeniable, but we’re still a long ways from anything that works well.

09:50 And the problem is with the fundamental nature of it doing the the congestion control at the sender side. It just doesn’t work very well. So, you end up with a lot of queue build-up. And in fact, you can see the only way to find out that there’s congestion is if there’s queues. And so, by that point we’re already experiencing delays. So, that’s a problem. Oops. There’s one other problem with TCP and RDMA also is that their their basic data model is a byte stream. Just a stream of bytes with no differentiation in it. So, if you send a series of messages say through a TCP socket, they get serialized into that

10:24 stream. And on this slide, I’ve you know, I’ve shown the messages appear like they have different colors in the stream. Well, there are no colors in real life. TCP has no idea where the message boundaries are. And that also makes life hard. For example, you don’t know how much more data is coming. If you knew how big the message was, you’d know how much more is coming. And you can’t prioritize short messages, which we’d really like to do, get the short messages through faster. And you can end up with what’s called head-of-line blocking, where somebody sends a series of messages to the same destination and they send two really large ones and then a small one

10:58 after that, they get stuck behind them in that stream. And so, it gets delayed. And again, you have tail latency issues. So, all in all, TCP and RDMA are just not well suited to this environment. So, what do we do? Well, what I’d like to do next is tell you about a new protocol called Homa that we’ve developed at Stanford, which was based on a completely clean slate redesign for network transport. If you could start from scratch and rethink how you do transport for data centers, how would you do it? And it turns out in

11:28 Homa, virtually every major design decision is different from TCP and RDMA. TCP for all the amazing things it’s done is just not a good match to today’s data centers, nor RDMA. So, what Homa does particularly well is to manage a combination of large and small messages and to make sure that the messages have rich short messages have really low latency. So, this started off as a PhD dissertation for one of my students, Ben Moazeni, and then the results were so great that I decided to make it my personal project to see if we can get it out of the lab

12:02 and into production. Uh as you may know, I’m I’m not like most professors in that I love to code, and so I turned this into my my own programming project. I created a kernel module for Linux. I’m currently working through the process of getting that upstreamed into the kernel. It’s available on GitHub for download. So, let me tell you just a little bit about how Homa works. I want to mention three things. First, it’s message-based, not stream-based. >> [clears throat] >> In fact, the fundamental unit at Homa is a remote procedure call, which consists of two things: a request message sent

12:35 from a client to a server, and then a response message returned back from the server to the client. So, the key thing here is that Homa knows about message lengths. They’re buried in the transport all the way down to the bottom, and this has a bunch of advantages. First, it allows us to predict the future. As soon as a receiver gets the first packet of a message, it knows exactly how much more data the sender wants to send, and that’s so much so much more information for doing congestion control.

13:05 Second, Homa prioritizes shorter messages. It uses SRPT, shortest remaining processing time first, to try and prioritize shorter messages. And third, because messages are all independent, they’re not serialized into a stream, every message is independent. Shorter messages can bypass long ones, so they don’t get queued behind long messages. The second thing about Homa that’s different is that it controls congestion from the receiver. Now, when you think about it, this makes

13:36 sense because the congestion happens primarily at that last down link to the receiver. And so, the receiver has way more information. In fact, with Homa, as soon as it gets the first packet of a message, it knows exactly how much more is coming. So, it has essentially complete information about congestion and it can therefore respond to congestion much more quickly and much more precisely. The way things work with Homa is that when a sender has a message to send, it breaks it up into packets, but it only

14:07 transmits the first few packets, those are called unscheduled packets, to the receiver. Packets after that are called scheduled packets, and they only get transmitted when the receiver asks for them. So, the receiver will send grant packets back. They’ll paste them out and send those back to the sender over time, telling the sender, “It’s now time for you to send me the next chunk of data.” And the receiver can delay those grants. So, for example, if the receiver has 10 messages that are incoming, there’s no point in sending grants to all 10 of

14:38 them because then you’ll just get congestion in the in the top of rack queues. So, we can use the grants to reduce congestion, and then it can also use the grants to give preference to its most favored messages, which would be the shorter ones. So, it’s a way of of implementing SRPT by favoring short messages. The third aspect of Homa is that it takes advantage of the priority queues in modern switches. So, modern data center switches have more than one queue at each egress port, typically eight, and they can be used in a priority

15:10 mechanism where packets get transmitted preferentially from the highest priority queue. So, I’ve shown only two queues on the slide here, but typically there’s more than that. You can specify in a packets, using the various fields of the packet, you can specify which queue it should go into. And so, Homa dynamically makes those choices in a way to give priority to shorter messages. So, if we go back to the incast example from a few slides ago, all of those long messages will pile up in the lowest priority queue. But, if there’s a short message coming, it will use a higher priority queue. And so, it will immediately bypass

15:44 all of the queued packets from the short from the the longer messages and get through to the destination more quickly. So, how much of a difference does this make? Uh here’s a uh this slide I’ve got one sample benchmark that I use as part of my tuning and evaluation of Homa. It consists of a workload of a bunch of machines on a network that are exchanging messages back and forth of different sizes, ranging from very small to very large. And on this graph, you can see on the x-axis is the message length, so from about 50 bytes up to a megabyte. The y-axis shows you the

16:18 round-trip time for messages of that length. So, this request this uses request and response messages that are the same length. You can see TCP in green, Homa in blue. And the y-axis is is round-trip time, so lower is better. And for each protocol, I’ve got two curves. One curve is the P50 curve, that’s the median latency for messages of this length. And then P99 is the 99th percentile, i.e. tail latency for messages of this length. So, I want to point out two things. First,

16:49 the P99 for short messages is dramatically better for Homa. So, with TCP, it’s more than a millisecond tail latency, Homa is less than 100 microseconds, about 13 times faster. Second, interestingly, you might think that because Homa favors shorter messages, that long messages suffer and get worse performance. It turns out that’s actually not the case. Even on the longest messages, Homa is almost a factor of two better than TCP. I don’t have time to explain that today, but it has to do with the fact that Homa

17:20 uses run-to-completion approaches, which are much more effective than fair than the fair scheduling used by TCP. So, just to wrap up, the role of short messages in AI appears to be increasing. I think I think it’s likely that it’s going to continue to increase. We’ll see over the next year or two if that happens. And I just want to pose a question to you, you know, as you’re running your applications and measuring performance and seeing what the bottlenecks are, ask yourself, is high latency for short messages affecting your throughput? If the answer is yes, then just know

17:54 there is a solution available. You should give Homa a try. You can probably reduce your tail latency by an order of magnitude or more. And by the way, this is Homa is basically my life mission right now. I’ve sort of semi-retired from Stanford, and the reason I did that is so I can spend 100% of my time hacking on Homa. So, I’d be delighted to work with you and help you if you decide you want to experiment with Homa. If you need help getting started, answer questions, bug fixes, whatever, you know, I’d be happy to work with you to try and make you successful with it. So, if that is interesting, feel free to contact me. My email’s on the slide, or you can Google

18:26 me, too, and find me over the internet. So, thanks very much for listening, and I hope to hear from some of you. >> [music]