Hi Vishal,
The e1000g man page is a bit out of date. Now the default value of
MaxNumReceivePackets is 128. This parameter actually does not affect
interrupt coalescing. It specifies the max packet number that the driver
can receive in one interrupt. If the packets in the rx descriptor ring
exceeds the number, some packets will be left to be processed when the
next interrupt comes. It does not coalesce any interrupt.
For rx interrupt coalescing, only the three parameters take effect:
intr_throttling_rate, or the combination of rx_intr_delay and
rx_intr_abs_delay. For more details about those parameters, you can
refer to the following document:
http://www.intel.com/design/network/applnots/ap450.htm
To observe the interrupt coalescing rate, you can use mpstat command
during the stress test to check the interrupt rate and estimate the
coalescing result.
Regards,
Ted
Hi Ted,
Thanks! Now, the e1000g document does not mention the default
interrupt throttling rate for opensolaris. I understand what you are saying
regarding the interrupt throttling rate, but I am very confused about this
whole thing. I read here
(http://docs.sun.com/app/docs/doc/819-2254/e1000g-7d?a=view) that for e1000g,
'MaxNumReceivePackets' defaults to 32, so it cannot handle more than 32 packets
per interrupt. It looks like there are far too many factors affecting the
interrupt coalescing. Is there a way to empirically observe the coalescing rate?
Thank you,
Vishal
_______________________________________________
networking-discuss mailing list
[email protected]
