[OpenIndiana-discuss] Crossbow performance & bit rot

[Jason Matthews]

From: Lou Picciano [mailto:loupicciano at comcast.net] 



> Yes, Jason, now that James has mentioned the 'Interrupt Storm' matter... 

 

> You've piqued my interest. Any chance you're having all these problems on 

> a SandyBridge system? Would love to hear more. In our case, we'd seen 

> interrupt faults rising into the _multiple millions_, over a 5-second 

> interval, before all hell started breaking loose. Though our (latest

> Series of) problems were not network-related, they may be responsible for

> problems with ZFS. 

 

 

Hi Lou,

 

It is an L5630 CPU, so no Sandy Bridge. My interrupts count totals between
16k-30k, normally, depending on network load. We some times see decent size
spikes on each web server where in bound http traffic increases to the tune
of 100 mb/s per box. This generally triggers about 150k pps/physical server.
I would imagine interrupts are near those levels as well. 

 

I haven't instrumented interrupt activity yet but it appears nominal to me.
An interrupt storm was the first thing I looked at about four weeks ago. I
just don't see any evidence of it. 

 

I don't have much going on in terms of zfs i/o. I am only logging proxy
errors to upstream servers. 

 

I found that about 8 megabits of traffic or less, my problem disappears. The
problem is at 8 megabits, I have a cost ineffective platform.

 

 

 

 

Thanks,

j.