RE: Performance Problems.. Server hardware smoked by $500 box?

From: John Straiton (jsmailing_at_clickcom.com)
Date: 09/12/03

  • Next message: Sean Kelly: "ipfw" 
    To: "'Mark Terribile'" <materribile@yahoo.com>, "'Marc Slemko'" <marcs@znep.com>
Date: Fri, 12 Sep 2003 10:30:58 -0400

    > Can you run a simple, processor/memory-intensive test
    > on the development machine and the Dell box?

    I ran ubench on both. Here's what the man page for it says:

             Ubench is executing rather senseless mathematical integer and
    floating-
           point calculations for 3 mins concurrently using several
    processes, and
           the result is Ubench CPU benchmark. The ratio of floating-point
    calcu-
           lations to integer is about 1:3. Ubench will spawn about 2
    concurrent
           processes for each CPU available on the system. This ensures all
    avail-
           able raw CPU horsepower is used.

           Ubench is executing rather senseless memory allocation and
    memory to
           memory copying operations for another 3 mins concurrently using
    several
           processes, and the result is Ubench MEM benchmark.

    Does that sound mean enough? I saw your function below but I know not
    the ways of C.

    > Can you do a raw test of performance on the NFS
    > mounts from the two systems? (Do they use the same
    > read size? Do they have the same readahead setting?)
    Both are mounted via /etc/fstab with default options as seen below (this
    is our php sessions mount):
    209.198.22.23:/var/sessions /sessions-on-db nfs rw 0 0

    > Are you using UDP mounts on both machines?
    Unless that's the default, no. I'd probably be of the opinion that since
    the mounts are all 2-way, that possible data loss from dropped UDP
    packets would be unacceptable.

    > Under load, what does the CPU line on systat -vmstat
    > look like on the two machines?

    I ran a test on both machines by running an abusing apache benchmark on
    them and then taking a snapshot of the report after 800+ requests had
    completed. This virtually brought the production machine to a halt while
    the development one kept putzin' along just fine. I don't know how to
    interpret the results tho as I've never seen this test before.

    During the height of an "ab -c 100 -n 1000" test against the machine,
    systat -vmstat reported this on the production webserver (sorry it's
    gonna be ugly)

        2 users Load 37.46 29.30 15.58 Sep 12 10:32

    Mem:KB REAL VIRTUAL VN PAGER SWAP
    PAGER
            Tot Share Tot Share Free in out in
    out
    Act 282112 9508 719504 15852 747888 count
    All 532264 16048 817988 27688 pages
                                                         1538 zfod
    Interrupts
    Proc:r p d s w Csw Trp Sys Int Sof Flt 138 cow 489
    total
        88 33 1470 1792 4087 905 141 1691 106688 wire
    stray 0
                                                       275748 act
    stray 1
    26.4%Sys 4.2%Intr 69.4%User 0.0%Nice 0.0%Idl 145200 inact
    stray 6
    | | | | | | | | | | cache
    npx0 13
    =============++>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>> 747888 free 1
    fxp0 11
                                                              daefr 260
    fxp1 10
    Namei Name-cache Dir-cache 247 prcfr
    ata0 14
        Calls hits % hits % react
    ahc0 5
        13160 13103 100 pdwak
    ahc1 3
                                                              pdpgs
    atkbd0 1
    Disks da0 acd0 fd0 pass0 pass1 intrn 100
    clk 0
    KB/t 0.00 0.00 0.00 0.00 0.00 114880 buf 128
    rtc 8
    tps 0 0 0 0 0 187 dirtybuf
    MB/s 0.00 0.00 0.00 0.00 0.00 86201 desiredvnodes
    % busy 0 0 0 0 0 21550 numvnodes
                                                        12593 freevnodes

    Doing the same thing against the development server gave this:

        1 users Load 23.00 8.48 3.50 Sep 12 10:26

    Mem:KB REAL VIRTUAL VN PAGER SWAP
    PAGER
            Tot Share Tot Share Free in out in
    out
    Act 330304 5192 580316 11540 48904 count
    All 985244 9320 1510820 19436 pages
     
    Interrupts
    Proc:r p d s w Csw Trp Sys Int Sof Flt 22 cow 1704
    total
         6 128 5748 726 2700 222 456 136576 wire
    stray 0
                                                       336964 act
    stray 6
    10.9%Sys 6.2%Intr 82.9%User 0.0%Nice 0.0%Idl 468416 inact
    stray 7
    | | | | | | | | | | 42536 cache
    npx0 13
    =====++++>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>> 6368 free 1464
    xl0 11
                                                              daefr
    rl0 12
    Namei Name-cache Dir-cache prcfr 12
    ata0 14
        Calls hits % hits % react
    fdc0 6
        15104 15040 100 pdwak 100
    clk 0
                                          428 zfod pdpgs 128
    rtc 8
    Disks ad0 224 ofod intrn
    KB/t 16.00 52 %slo-z 110928 buf
    tps 12 84 tfree 192 dirtybuf
    MB/s 0.18 68139 desiredvnodes
    % busy 0 37187 numvnodes
                                                         5864 freevnodes

    John Straiton jks@ clickcom.com Clickcom, Inc 704-365-9970x101

    _______________________________________________
    freebsd-performance@freebsd.org mailing list
    http://lists.freebsd.org/mailman/listinfo/freebsd-performance
    To unsubscribe, send any mail to "freebsd-performance-unsubscribe@freebsd.org"

  • Next message: Sean Kelly: "ipfw"