Subject: Re: igb - no copybreak used - msg#00103

Previous Message by Date: click to view message preview

Re: igb - very high rate of interrupts when usingmultiple rx queues

As per David Miller, the networking stack maintainer, and Jamal Hadi, the creator of NAPI, hardware mitigation should be used to limit the total number of interrupts per second generated in the case of fast cpus and packet rates/sizes that take less time to process than to receive the packet. see http://kernel.org/pub/linux/kernel/people/hadi/docs/UKUUG2005.pdf http://marc.info/?l=linux-netdev&w=2&r=1&s=possible+napi+improvements&q= b ________________________________ From: Don Ash [mailto:donash4@xxxxxxxxx] Sent: Monday, January 21, 2008 3:15 PM To: Brandeburg, Jesse Cc: e1000-devel@xxxxxxxxxxxxxxxxxxxxx Subject: Re: [E1000-devel] igb - very high rate of interrupts when usingmultiple rx queues How does InterruptThrottleRate or interrupt moderation code work together with NAPI polling scheme which also reduces interrupts rate? On Jan 22, 2008 12:24 AM, Brandeburg, Jesse <jesse.brandeburg@xxxxxxxxx> wrote: you can limit the interrupt rate (to 20000) by specifying insmod igb.ko InterruptThrottleRate=20000,20000 Our interrupt moderation code currently favors unconditional low latency for small packets, when it should probably be reducing the interrupt rate for line rate small packets. Jesse -----Original Message----- From: e1000-devel-bounces@xxxxxxxxxxxxxxxxxxxxx [mailto: e1000-devel-bounces@xxxxxxxxxxxxxxxxxxxxx <mailto:e1000-devel-bounces@xxxxxxxxxxxxxxxxxxxxx> ] On Behalf Of Don Ash Sent: Monday, January 21, 2008 2:21 PM To: e1000-devel@xxxxxxxxxxxxxxxxxxxxx Subject: [E1000-devel] igb - very high rate of interrupts when usingmultiple rx queues i am testing ip forwarding packet rate with RHEL 5 using 1 core only. on igb 1.12.2.4 <http://1.12.2.4/> default configuration there is around 20k interrupts/sec with decent packer rate but when i use IntMode=3 (4 rx queues) and all queues are being fed there are around 300k uinterrupts/sec and thus the packet rate drops significantly although it should stay the same since the "only" difference is that there are 4 queus (assigned to 1 core) being fed instead of 1 but i guess the igb driver is not tuned ok to handle this case. ------------------------------------------------------------------------- This SF.net email is sponsored by: Microsoft Defy all challenges. Microsoft(R) Visual Studio 2008. http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/_______________________________________________ E1000-devel mailing list E1000-devel@xxxxxxxxxxxxxxxxxxxxx https://lists.sourceforge.net/lists/listinfo/e1000-devel

Next Message by Date: click to view message preview

Re: igb - no copybreak used

So, will it be removed also from e1000 driver soon? On Jan 22, 2008 2:07 AM, Brandeburg, Jesse <jesse.brandeburg@xxxxxxxxx> wrote: > because the gain was marginal, and the method has fallen out of favor. > > routing packets is hurt in the default case by having copybreak enabled. > > > > -----Original Message----- > From: e1000-devel-bounces@xxxxxxxxxxxxxxxxxxxxx > [mailto:e1000-devel-bounces@xxxxxxxxxxxxxxxxxxxxx] On Behalf Of Don Ash > Sent: Monday, January 21, 2008 3:38 PM > To: e1000-devel@xxxxxxxxxxxxxxxxxxxxx > Subject: [E1000-devel] igb - no copybreak used > > Why doesn't the igb driver make use of the "copybreak" which is used by > e1000? > ------------------------------------------------------------------------- This SF.net email is sponsored by: Microsoft Defy all challenges. Microsoft(R) Visual Studio 2008. http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/_______________________________________________ E1000-devel mailing list E1000-devel@xxxxxxxxxxxxxxxxxxxxx https://lists.sourceforge.net/lists/listinfo/e1000-devel

Previous Message by Thread: click to view message preview

igb - no copybreak used

Why doesn't the igb driver make use of the "copybreak" which is used by e1000? ------------------------------------------------------------------------- This SF.net email is sponsored by: Microsoft Defy all challenges. Microsoft(R) Visual Studio 2008. http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/_______________________________________________ E1000-devel mailing list E1000-devel@xxxxxxxxxxxxxxxxxxxxx https://lists.sourceforge.net/lists/listinfo/e1000-devel

Next Message by Thread: click to view message preview

Re: igb - no copybreak used

So, will it be removed also from e1000 driver soon? On Jan 22, 2008 2:07 AM, Brandeburg, Jesse <jesse.brandeburg@xxxxxxxxx> wrote: > because the gain was marginal, and the method has fallen out of favor. > > routing packets is hurt in the default case by having copybreak enabled. > > > > -----Original Message----- > From: e1000-devel-bounces@xxxxxxxxxxxxxxxxxxxxx > [mailto:e1000-devel-bounces@xxxxxxxxxxxxxxxxxxxxx] On Behalf Of Don Ash > Sent: Monday, January 21, 2008 3:38 PM > To: e1000-devel@xxxxxxxxxxxxxxxxxxxxx > Subject: [E1000-devel] igb - no copybreak used > > Why doesn't the igb driver make use of the "copybreak" which is used by > e1000? > ------------------------------------------------------------------------- This SF.net email is sponsored by: Microsoft Defy all challenges. Microsoft(R) Visual Studio 2008. http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/_______________________________________________ E1000-devel mailing list E1000-devel@xxxxxxxxxxxxxxxxxxxxx https://lists.sourceforge.net/lists/listinfo/e1000-devel