Date: Wed, 29 Oct 2008 13:37:44 +0100 From: Bartosz Giza <gizmen@blurp.pl> To: Alexander Motin <mav@freebsd.org> Cc: freebsd-net@freebsd.org Subject: Re: two NIC on 2 core system (scheduling problem) Message-ID: <200810291337.44899.gizmen@blurp.pl> In-Reply-To: <49083CBD.1000701@FreeBSD.org> References: <1225203780.00029971.1225190402@10.7.7.3> <200810290953.28237.gizmen@blurp.pl> <49083CBD.1000701@FreeBSD.org>
next in thread | previous in thread | raw e-mail | index | archive | help
Wednesday 29 of October 2008 11:36:45 Alexander Motin napisa=C5=82(a): > Bartosz Giza wrote: > > Tuesday 28 of October 2008 19:10:43 Alexander Motin napisa=C5=82(a): > >> Bartosz Giza wrote: > >>>> The CPU time you see there includes much more then just a card > >>>> handling itself. It also includes CPU time of the most parts of > >>>> network stack used to process received packet. So if you have NAT, > >>>> big firewall, netgraph or any other CPU-hungry actions done with > >>>> packets incoming via em0 you will see such results. > >>>> Even more interesting is that if bge0 or fxp0 cards will require > >>>> much CPU time to send packet, this time will also be accounted to > >>>> em0 process. > > > > I have checked this and you are right. When i turned off ipfw; taskq > > process started to use less cpu. But still what is strange why > > processing from other cards are counted in em0 taskq ? > > What you mean by "processing from other cards"? em0 taskq counts all > processing caused by packets incoming via em0 up to and including > processing of their transmission by bge/fxp drivers. Same is about > bge/fxp. If bge/fxp/em drivers would have separate transmission > processes - you would see them, but they don't, so their CPU time > accounted to the caller. Ok now i think understand this. If packet enters via em0 all processing of= =20 packet filters(and others) way up to sending packet via other interface is= =20 counted to em0 taskq( even overhead of packet filter when packet leaves the= =20 interface) ? Basicly overhead from passing packet to packet filter twice=20 (on in and out) > > This is quite strange and in that > > way em0 taskq process is using more cpu on one of the cores. So what i > > think the best would be to have only em NICs because processing of the > > packets would be splitted to those taskq processes is that right ? > > em0 processes packets in separate process named taskq, bge does it > directly in interrupt handler process. There is no principal difference > for you I think. So now i am lost again. If packet filtering on bge card is counted to irq17= :=20 bge0 process so i think it should use more cpu. =46rom what you wrote there should be no difference for me if card use tas= q=20 or irq. Those processes do exactly the same thing? If that is true so why=20 there is so much difference in cpu usage: 20 root 1 -68 - 0K 8K - 0 161:01 18.75% em0 taskq 21 root 1 -68 - 0K 8K WAIT 1 100:10 5.47% irq17: bge0 23 root 1 -68 - 0K 8K WAIT 0 75:31 2.98% irq16: fxp1 If what you wrote is true that overhead of incomming packet on bge0 should= =20 be counted to irq17: bge0 So don't understand why there is so big cpu usage on em0. From what you are= =20 saying irq17 and em0 taskq should have similar usage. Even more bge0 passes= =20 about two times more traffic than em0. I simply don't understand this. > > Ok, good to know. But how is counted firewall overhead when i would > > have only bge cards. They don't use taskq so i assume i would see this > > as system usage correct ? > > You would see a lot of interrupt time in this case.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200810291337.44899.gizmen>