Re:A Tour Inside CloudFlare's Latest Generation Servers
by virtuallynathan on Monday July 22, 2013 @02:51PM
(#6085720)
If CloudFlare is using Kernel version 3.3 or higher, they should look into using fq_codel as their scheduler instead of pfifo_fast to decrease latency under load. I suspect the 16MB buffer in the NIC isn't doing them any favors.
[ Parent
| on HN ]
Re:A Tour Inside CloudFlare's Latest Generation Servers
by eastdakota on Monday July 22, 2013 @03:45PM
(#6086051)
I asked one of our kernel guys. Here's his response:
"We are indeed looking at Codel. We were actually working on backporting BQL+Codel to the 2.6.x kernel but the Google guys finally got the network stack under control enough for us to deploy >3.3. The 16MB of buffers hasn't hurt us much yet, and may in the long run save us from switches that have too shallow a buffer for the high contention ratios we run on the switch." -LinuXY
Re:A Tour Inside CloudFlare's Latest Generation Servers
by virtuallynathan on Monday July 22, 2013 @04:32PM
(#6086286)
Interesting, thanks for the response!
Re:A Tour Inside CloudFlare's Latest Generation Servers
by newman314 on Monday July 22, 2013 @05:40PM
(#6086639)
What patches from Google specifically?
Re:A Tour Inside CloudFlare's Latest Generation Servers
by croikle on Monday July 22, 2013 @03:12PM
(#6085851)
To be fair, at 10 gigabits that buffer is only 12.5 ms, but given their dedication to latency I can see it being worth investigating.
Re:A Tour Inside CloudFlare's Latest Generation Servers
by virtuallynathan on Monday July 22, 2013 @03:21PM
(#6085896)
Buffer Bloat is a real thing, it has to do with how the queuing algorithms work - I'm not the best at explaining it. I think it's mainly an issue in home routers / cable modems - but it cant hurt to do some Active Queue Management everywhere!