Goal: Enable multiple receive queues so that my servers can scale  
beyond what a single can handle in terms of interrupts (hard and soft).

Environment: ET dual port PCI-express cards (82576), CentOS 5.2 with  
2.6.18-92.1.22.el5 kernel, Dell PowerEdge 1950 servers, nginx  
webserver/proxies, igb driver 1.3.8.6

Current igb options: IntMode=3,3 InterruptThrottleRate=60000,60000  
LLIPort=80,80 LLIPush=1,1 LLISize=768,768

MSI-X: enabled according to /proc/interrupts (4 receive and 1 sending  
queue)

What I'm seeing: All of the interrupts are landing on CPU0 and using  
up too much time for my liking. I'd like to spread the interrupts  
around some more, isn't this what having multiple queues is for? The  
userland irqbalance is disabled for now, as I thought the hardware and  
driver were supposed to distribute things.

Is there anything I'm missing here? Do I need a more recent kernel for  
this to work properly?

Thanks.

-richard

------------------------------------------------------------------------------
_______________________________________________
E1000-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/e1000-devel

Reply via email to