device polling takes more CPU hits??

From: James (haesu_at_towardex.com)
Date: 07/26/04

  • Next message: Don Bowman: "RE: device polling takes more CPU hits??" 
    Date: Mon, 26 Jul 2004 10:42:01 -0400
To: freebsd-net@freebsd.org

    Hi all,

    I've got a weird case on my hands on a 2.8ghz xeon w/ HTT working as a router w/
    device polling... This is a single-processor 2.8ghz xeon, not dual/quad, etc.
    The kernel does have SMP compiled in though.

    The box has two gig-e cards, em0 and bge0. bge0 is the uplink to the core,
    em0 is the downlink to the ethernet switch with 802.1q vlans for customer
    aggregation. During daytime, the box pushes about 15kpps at rate of roughly
    12% interrupt CPU load. Sometimes when it spikes to 100kpps (rare, but happens),
    cpu load goes up as high as 30% on interrupts.

    Now these figures are w/ device polling off. As soon as I turn device polling
    ON, interrupt load climbs from 12% to 23%. During spikes, it climbs to about
    48% to 50%.

    Any idea why device polling is kind of having... negative impact? Is this b/c
    I have SMP compiled on a box that really doesn't have two cpu's?? Is SMP+APIC_IO
    support even required for HTT use?

    The version btw is FreeBSD 4.9-STABLE.

    hardware info:
    Jul 21 23:09:24 r2.bos /kernel: CPU: Intel(R) Xeon(TM) CPU 2.80GHz (2788.16-MHz 686-class CPU)
    Jul 21 23:09:24 r2.bos /kernel: Origin = "GenuineIntel" Id = 0xf29 Stepping = 9
    Jul 21 23:09:25 r2.bos /kernel: em0: <Intel(R) PRO/1000 Network Connection, Version - 1.7.25> port 0xccc0-0xccff mem 0xfcd00000-0xfcd3ffff,0xfcd40000-0xfcd5ffff irq 7 at device 6.0 on pci3
    Jul 21 23:09:25 r2.bos /kernel: bge0: <Broadcom BCM5704C Dual Gigabit Ethernet, ASIC rev. 0x2002> mem 0xfcf20000-0xfcf2ffff,0xfcf30000-0xfcf3ffff irq 2 at device 0.0 on pci2
    Jul 21 23:09:25 r2.bos /kernel: bge1: <Broadcom BCM5704C Dual Gigabit Ethernet, ASIC rev. 0x2002> mem 0xfcf00000-0xfcf0ffff,0xfcf10000-0xfcf1ffff irq 5 at device 0.1 on pci2

    Thanks for any tips!
    -J 

    -- 
James Jun                                            TowardEX Technologies, Inc.
Technical Lead                        Network Design, Consulting, IT Outsourcing
james@towardex.com                  Boston-based Colocation & Bandwidth Services
cell: 1(978)-394-2867           web: http://www.towardex.com , noc: www.twdx.net
_______________________________________________
freebsd-net@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-net
To unsubscribe, send any mail to "freebsd-net-unsubscribe@freebsd.org"
  • Next message: Don Bowman: "RE: device polling takes more CPU hits??"

    Relevant Pages

    • Re: Interrupts + Polling mode (similar to Linuxs NAPI)
      ... The advantage of having interrupts is faster response ... oh, I was under the impression that polling spun in a tight loop, thus ... I assume it will get to 100% CPU load if I saturate my network. ... the complete definition should be "no more packets for X seconds". ...
      (freebsd-net)
    • Re: routing for 1000 users and 10Mbit internet.
      ... that if you try polling and it ... CPU load, ... if you absolutely want to come up with a reason to buy new ... hardware, this is as good a reason as any. ...
      (freebsd-questions)
    • Re[2]: Interrupts + Polling mode (similar to Linuxs NAPI)
      ... Interrupts + Polling mode ... I assume it will get to 100% CPU load if I saturate my network. ... the complete definition should be "no more packets for X seconds". ...
      (freebsd-net)
    • Re: Interrupts + Polling mode (similar to Linuxs NAPI)
      ... The advantage of having interrupts is faster response ... oh, I was under the impression that polling spun in a tight loop, thus ... I assume it will get to 100% CPU load if I saturate my network. ... the SMP kernels we have now (in other words: ...
      (freebsd-net)