VERY slow performance on igb+FreeBSD8.2+mpd5.6

[Коньков Евгений]

Здравствуйте, Коньков.

Вы писали 23 февраля 2012 г., 9:19:45:


КЕ> #uname FreeBSD 8.3-PRERELEASE #2 r231881: Thu Feb 23 00:53:28 UTC 2012
КЕ> и Version 5.6 (root@ 10:03 20-Feb-2012)
КЕ> http://www.speedtest.net/result/1790445113.png
КЕ> try to reconnect to mpd 10-20times and you get next:
КЕ> http://www.speedtest.net/result/1790454801.png

КЕ>  mpd has no any sessions except this one.

КЕ> net.graph.recvspace=524288
КЕ> net.graph.maxdgram=524288

КЕ> net.inet.ip.dummynet.io_fast=1
КЕ> net.inet.ip.dummynet.pipe_slot_limit=1000

КЕ> dev.igb.0.rx_processing_limit=4096
КЕ> dev.igb.1.rx_processing_limit=4096
КЕ> dev.igb.2.rx_processing_limit=4096
КЕ> dev.igb.3.rx_processing_limit=4096

КЕ> net.route.netisr_maxqlen=4096

КЕ> hw.igb.rxd=4096
КЕ> hw.igb.txd=4096
КЕ> hw.igb.max_interrupt_rate=32000

КЕ> net.isr.defaultqlimit=4096
КЕ> net.link.ifqmaxlen=10240

КЕ> net.graph.maxdata=32768
КЕ> net.graph.maxalloc=32768

КЕ> net.inet.tcp.tcbhashsize=4096         # tcb hash size
КЕ> net.inet.tcp.syncache.hashsize=1024   # syncache hash size
КЕ> net.inet.tcp.syncache.bucketlimit=100 # syncache bucket limit
КЕ> net.isr.bindthreads=0                 # do not bind threads to CPUs
КЕ> net.isr.direct=0                      # interrupt handling via multiple CPU
КЕ> net.isr.direct_force=0                # "
КЕ> net.isr.maxthreads=4                  # Max number of threads for
КЕ> NIC IRQ balancing (4 cores in box)

КЕ> vmstat -z | grep Gra
КЕ> ITEM                     SIZE     LIMIT      USED      FREE  REQUESTS  FAILURES
КЕ> NetGraph items:            72,    32770,        0,      377,   143887,        0
КЕ> NetGraph data items:       72,    32770,        0,      435,   241884,        0

Can any one help to debug that problem?

-- 
С уважением,
 Коньков                          mailto:kes-kes at yandex.ru