News for hackers. Stuff that matters.
Re:The Ethernet PAUSE frame by greglindahl on Monday August 22, 2016 @09:29PM (#12340711)
Uh, no. Alltoall is a challenge for MPI, but not for the reason you describe. TCP windows mean that the receivers aren't the problem. It's all the switch queues in the middle.
Re:The Ethernet PAUSE frame by dekhn on Monday August 22, 2016 @10:11PM (#12340898)
welp, I measured the problem on my machines, and enabling pause frames on the switches fixed the problem...
Re:The Ethernet PAUSE frame by vardump on Tuesday August 23, 2016 @02:25AM (#12341793)
TCP windows won't save you. TCP has no way to magically know when some buffer is full. Instead it notices packet loss and interprets it as congestion. Which is not what you want, because it can significantly reduce throughput.
Re:The Ethernet PAUSE frame by r4um on Tuesday August 23, 2016 @08:19AM (#12343068)
Yes it does, receiver side advertises window as 0 and persist timer kicks in. The send call then blocks until window recovers.
Re:The Ethernet PAUSE frame by vardump on Tuesday August 23, 2016 @11:01AM (#12344243)
That's the problem when you don't really even have any congestion, but just very high packet loss caused by small buffers. Transfer rate drops to nothing.

With pause frames you can avoid that situation.

Re:The Ethernet PAUSE frame by bcook on Tuesday August 23, 2016 @05:50AM (#12342492)
There's many types of TCP congestion algos and not all of them require dropped packets to begin the congestion avoidance.
Re:The Ethernet PAUSE frame by dekhn on Tuesday August 23, 2016 @09:33AM (#12343546)
In fact, at the time, I was working at LBL in a lab that developed new congestion avoidance algorithms. The problem was that most of our tuning for was for long-range links, while I was using the cluster for local high performance networking.