(no subject)

[ewx]

One of my more frustrating days at work. I spent most of the day tracking down a performance problem; once we worked out what we thought it was the fix seemed reasonably clear - but it doesn't seem to work in practice. So did I get the fix wrong, did we reach the wrong conclusion about the problem, or is the underlying OS broken in this area? And some extra work has appeared at short notice and with a rather short deadline.

Went swimming after work, the remains of my cold are still curtailing that activity somewhat though.

Comments

[sparrowsion]

Recent experience suggests that you might also want to investigate whether the compiler is contributing anything (positive or negative) to the issue. (We've got a situation where one of our performance-critical libraries gives equally good performance built with gcc or icl, but we know they're optimising in different areas....)

[ewx.livejournal.com]

It's interactions between TCP stacks that are causing us trouble - there's a 30ms timeout clearly visible in tcpdump and at higher layers - and the socket option that's supposed to change what we believe to be the relevant behaviour doesn't seem to have any effect. (And tracing and debugging are unusually effort-intensive on the target.)

[sparrowsion]

If it is a Nagle/delayed-ACK interaction, be glad that you're not doing it on Windows where you'd have a 200ms timeout....

[ewx.livejournal.com]

We are doing it on Windows as well l-(

[keirf.livejournal.com]

Cold + swimming = earch. Even when well I find the chlorine in the pool gives me a sore throat; when I start with a sore throat it becomes unbearable.