Menu
▾
▴
Re: [uml-devel] UML sleeping 5X too long? was Re: slart account (fwd)
|
From: William S. <wst...@po...> - 2003-01-11 00:25:29
|
Good day, all, Two quick notes; this problem _seems_ to have shown up when I put in 2.4.19-uml45, which is the first skas enabled kernel on zaphod. Secondly, David Coulson was kind enough to redo the test (on the same kernel) on his debian "testing" distribution; his came back with "real=18 seconds" David's versions: libc6 2.2.5-11.2 GNU C Library: Shared libraries and Timezone netkit-ping 0.10-9 The ping utility from netkit Cheers, - Bill On Fri, 10 Jan 2003, William Stearns wrote: > Dan runs a virtual machine on Zaphod. He and I both ran "time > ping -c 10 some_ip" on our respective virtual machines, eash running > 2.4.19-uml45. His is debian 30, mine is redhat 7.3. > My ping takes 10 seconds; his takes 60 seconds. > ? > The reason I'm bringing this up here is that while it appears to > be some library related issue (Dan, which glibc and ping tool are you > using, and what versions are they?), it sounds like it's somehow related > to the fact that UML divides the available CPU time by (5?) to figure out > how much the VM can take. > Any ideas? > Cheers, > - Bill > > For reference, slart is a rh73: > [root@slartibartfast root]# time ping -c 10 anime.net > PING anime.net (208.8.184.120) from 66.59.111.182 : 56(84) bytes of data. > 64 bytes from sasami.anime.net (208.8.184.120): icmp_seq=0 ttl=44 time=104.635 msec > 64 bytes from sasami.anime.net (208.8.184.120): icmp_seq=1 ttl=44 time=108.821 msec > 64 bytes from sasami.anime.net (208.8.184.120): icmp_seq=2 ttl=44 time=97.552 msec > 64 bytes from sasami.anime.net (208.8.184.120): icmp_seq=3 ttl=44 time=98.327 msec > 64 bytes from sasami.anime.net (208.8.184.120): icmp_seq=4 ttl=44 time=131.072 msec > 64 bytes from sasami.anime.net (208.8.184.120): icmp_seq=5 ttl=44 time=137.603 msec > 64 bytes from sasami.anime.net (208.8.184.120): icmp_seq=6 ttl=44 time=103.906 msec > 64 bytes from sasami.anime.net (208.8.184.120): icmp_seq=7 ttl=44 time=109.415 msec > 64 bytes from sasami.anime.net (208.8.184.120): icmp_seq=8 ttl=44 time=128.864 msec > 64 bytes from sasami.anime.net (208.8.184.120): icmp_seq=9 ttl=44 time=227.517 msec > > --- anime.net ping statistics --- > 10 packets transmitted, 10 packets received, 0% packet loss > round-trip min/avg/max/mdev = 97.552/124.771/227.517/36.809 ms > > real 0m10.407s > user 0m0.000s > sys 0m0.000s > > [root@slartibartfast root]# rpm -q glibc iputils > glibc-2.2.4-31 > iputils-20001110-6 > > > ---------- Forwarded message ---------- > Date: Fri, 10 Jan 2003 15:57:58 -0800 (PST) > From: Dan Hollis <go...@an...> > To: William Stearns <wst...@po...> > Subject: Re: slart account (fwd) > > On Fri, 10 Jan 2003, William Stearns wrote: > > --- www.yahoo.akadns.net ping statistics --- > > 10 packets transmitted, 10 packets received, 0% packet loss > > round-trip min/avg/max/mdev = 13.626/27.870/48.083/10.704 ms > > > > real 0m9.568s > > user 0m0.000s > > sys 0m0.000s > > > > The "real" line indicates that it took 9 and a half seconds of > > wall clock time. How much time did yours take in actual wall clock time? > > bash-2.05a# time ping -c 10 anime.net > PING anime.net (208.8.184.120) from 66.59.111.169 : 56(84) bytes of data. > 64 bytes from sasami.anime.net (208.8.184.120): icmp_seq=1 ttl=44 time=106 ms > 64 bytes from sasami.anime.net (208.8.184.120): icmp_seq=2 ttl=44 time=178 ms > 64 bytes from sasami.anime.net (208.8.184.120): icmp_seq=3 ttl=44 time=138 ms > 64 bytes from sasami.anime.net (208.8.184.120): icmp_seq=4 ttl=44 time=138 ms > 64 bytes from sasami.anime.net (208.8.184.120): icmp_seq=5 ttl=44 time=116 ms > 64 bytes from sasami.anime.net (208.8.184.120): icmp_seq=6 ttl=44 time=124 ms > 64 bytes from sasami.anime.net (208.8.184.120): icmp_seq=7 ttl=44 time=108 ms > 64 bytes from sasami.anime.net (208.8.184.120): icmp_seq=8 ttl=44 time=143 ms > 64 bytes from sasami.anime.net (208.8.184.120): icmp_seq=9 ttl=44 time=131 ms > 64 bytes from sasami.anime.net (208.8.184.120): icmp_seq=10 ttl=44 time=157 ms > > --- anime.net ping statistics --- > 10 packets transmitted, 10 received, 0% loss, time 48371ms > rtt min/avg/max/mdev = 106.561/134.468/178.353/21.122 ms > > real 0m59.003s > user 0m0.000s > sys 0m0.000s > > each ping takes about 5 seconds in between each line, but the roundtrip > time is calculated properly. this tells me that something is really broken > with the kernel sleep() calls. > > interactive performance seems ok, but anything which has to do timed > sleeps runs 1/6 normal speed. which means this *kills* the game server. > > -Dan > -- > [-] Omae no subete no kichi wa ore no mono da. [-] --------------------------------------------------------------------------- "Tatu Ylonen was given public awards & wined & dined by the Finnish government for making ssh available to the world. Phil Zimmerman got a prison-term-commuted-to-community-service for doing the same with PGP. There is **NO** doubt that US govermnmet laws, PAST and PRESENT, HAVE hindered the development, commercialization and sales of cryptographic software." -- Duncan Napier <na...@na...> -------------------------------------------------------------------------- William Stearns (wst...@po...). Mason, Buildkernel, freedups, p0f, rsync-backup, ssh-keyinstall, dns-check, more at: http://www.stearns.org Linux articles at: http://www.opensourcedigest.com -------------------------------------------------------------------------- |
