On Thu, Oct 24, 2013 at 2:24 PM, k simon <[email protected]> wrote: > Hi, Ziehau > > Yes, you are right. Not only netisr_cpu 2, and netisr_cpu 0-7 else. > > And my box’s version > # uname -a > DragonFly sq-l1-n2-df 3.4-RELEASE DragonFly v3.4.3.16.g5011f5-RELEASE #0: > Wed Oct 23 16:44:28 CST 2013 > root@sq-l1-n2-df:/usr/obj/usr/src/sys/MYKERNEL > x86_64 > > As a high load for IO and network load system, can I tuning something > for make it better? > > As for the TCP network load, you probably want to try the latest master, on which I have improved connect(2) performance greatly (both blocking and non-blocking). And Dillon also has done great amount of work to improve system overall performance in the latest master. One thing need to note is that you probably want to search in the mail archive for the 3.4 -> master upgrading information, since master now defaults to dports and libc has gone through significant changes.
Best Regards, sephe > > > # cat /boot/loader.conf > kern.ipc.nmbclusters="262144" > kern.ipc.maxsockets="204800" > ## tuning for tcp > net.inet.tcp.tcbhashsize="262144" > net.inet.tcp.reass.maxsegments="16384" > net.inet.tcp.syncache.hashsize=32768 > net.inet.tcp.syncache.bucketlimit=32 > net.inet.tcp.syncache.cachelimit=1048576 > > > # cat /etc/sysctl.conf > kern.maxfilesperproc=500058 > kern.ipc.somaxconn=65532 > net.inet.ip.portrange.first=10001 > net.inet.ip.portrange.last=65535 > net.inet.tcp.msl=5000 > net.inet.tcp.syncookies=0 > net.inet.tcp.sendspace=32768 > net.inet.tcp.recvspace=32768 > net.inet.tcp.recvbuf_auto=0 > net.inet.tcp.sendbuf_auto=0 > net.inet.tcp.tso=0 > kern.metadelay=3 > kern.dirdelay=4 > kern.filedelay=5 > vfs.write_behind=0 > vfs.ufs.dirhash_maxmem=16777216 > > > > Regards > Simon > > > > > > On 2013年10月24日, at 下午12:12, Sepherosa Ziehau <[email protected]> wrote: > > > > > On Thu, Oct 24, 2013 at 11:03 AM, k simon <[email protected]> wrote: > >> HI,list: >> >> I noticed when I exec "top -S" ,some process's state display >> "waitport" or "waitmsg", >> eg: >> netisr_cpu 2 waitport >> netisr_cpu 2 waitport >> haproxy waitmsg >> squid waitmsg >> >> What's this mean? Does it work normally? >> >> > 2 "netisr_cpu 2"? I believe there should only be one "netisr_cpu 2". > waitport is normal for netisr. If you are using 3.4 or if squid uses > blocking connect(2), waitmsg is normal then. > > Best Regards, > sephe > > >> Regards >> Simon >> >> >> > > > -- > Tomorrow Will Never Die > > > -- Tomorrow Will Never Die
