Re: kern.hz = 10
Ivan Voras writes: > 2009/1/30 Julian Stecklina : >> Mister Olli writes: >> >>> Hi, >>> >>> I've never experimented with the kern.hz variable, but on all >>> configuration examples for 8-CURRENT if always seen 'kern.hz=100'. >>> From where did you get your configuration? >> >> I use the default config (albeit renamed to BLITZXEN) from 8-CURRENT. > > Due to bad performance of high HZ in VMs, in 8-CURRENT it is > explicitly reduced to 10 if the kernel is running under any kind of > virtual environment. This is a somewhat theoretical number and > probably needs tuning. Probably. > If you're interested in trying it out, you can increase it by adding > "kern.hz=xxx" to /boot/loader.conf and seeing what is the actual > performance influence of high HZ under Xen. I am going to do that once the timing stuff is figured out. No point in making benchmarks when the clock is confused. :) I cannot repeat it enough: Still great to see FreeBSD on Xen! Regards, -- Julian Stecklina Well, take it from an old hand: the only reason it would be easier to program in C is that you can't easily express complex problems in C, so you don't. - Erik Naggum (in comp.lang.lisp) ___ freebsd-xen@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-xen To unsubscribe, send any mail to "freebsd-xen-unsubscr...@freebsd.org"
Re: kern.hz = 10
2009/1/30 Mister Olli : > I've run on 100HZ quite some time now and did have the impression that > 8-CURRENT is slow. It's even faster on copy processes as 7 under VMware > on a host machine which is 3 years younger... This might or (probably) might not be due to HZ. If you're interested in testing, here's what you should do: 1. compile a 8-CURRENT kernel without debugging (WITNESS, INVARIANTS & their support kernel options) 2. compile a 8-CURRENT world without malloc debugging (see http://wiki.freebsd.org/DefaultDebuggingKnobs) 3. run some repeatable tests - I'd suggest some file system benchmarks on a RAM (md) drive, see http://man.freebsd.org/md like bonnie++ and blogbench and some network tests with iperf 4. change HZ in loader.conf and test again, in exactly the same way as before (in 3.) Benchmarks that are not repeatable are useless. Repeatable means by you (so e.g. the host machine must be in the same state - no additional programs running, etc., see http://wiki.freebsd.org/BenchmarkAdvice) and by others when following your steps exactly. ___ freebsd-xen@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-xen To unsubscribe, send any mail to "freebsd-xen-unsubscr...@freebsd.org"
Re: kern.hz = 10
Hi, > Due to bad performance of high HZ in VMs, in 8-CURRENT it is > explicitly reduced to 10 if the kernel is running under any kind of > virtual environment. This is a somewhat theoretical number and > probably needs tuning. Oh I didn't knew that. *SORRY* > If you're interested in trying it out, you can increase it by adding > "kern.hz=xxx" to /boot/loader.conf and seeing what is the actual > performance influence of high HZ under Xen. I've run on 100HZ quite some time now and did have the impression that 8-CURRENT is slow. It's even faster on copy processes as 7 under VMware on a host machine which is 3 years younger... .. Mr. Olli ___ freebsd-xen@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-xen To unsubscribe, send any mail to "freebsd-xen-unsubscr...@freebsd.org"
Re: kern.hz = 10
2009/1/30 Julian Stecklina : > Mister Olli writes: > >> Hi, >> >> I've never experimented with the kern.hz variable, but on all >> configuration examples for 8-CURRENT if always seen 'kern.hz=100'. >> >>>From where did you get your configuration? > > I use the default config (albeit renamed to BLITZXEN) from 8-CURRENT. Due to bad performance of high HZ in VMs, in 8-CURRENT it is explicitly reduced to 10 if the kernel is running under any kind of virtual environment. This is a somewhat theoretical number and probably needs tuning. If you're interested in trying it out, you can increase it by adding "kern.hz=xxx" to /boot/loader.conf and seeing what is the actual performance influence of high HZ under Xen. ___ freebsd-xen@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-xen To unsubscribe, send any mail to "freebsd-xen-unsubscr...@freebsd.org"
Re: kern.hz = 10
On Thu, Jan 29, 2009 at 5:08 PM, Julian Stecklina wrote: > Hello, > > I have an 8-CURRENT (as of yesterday) DomU running. kern.hz is set to > 10. Is this intended? This gives some strange effects: > > $ ping www.google.de > PING www.l.google.com (74.125.39.99): 56 data bytes > 64 bytes from 74.125.39.99: icmp_seq=0 ttl=246 time=100.000 ms > 64 bytes from 74.125.39.99: icmp_seq=1 ttl=246 time=0.000 ms > 64 bytes from 74.125.39.99: icmp_seq=2 ttl=246 time=0.000 ms > It probably took a few ticks for ARP resolution to occur. -Kip > dmesg: > FreeBSD is a registered trademark of The FreeBSD Foundation. > FreeBSD 8.0-CURRENT #0 r187877: Thu Jan 29 16:32:39 CET 2009 >jul...@bsd-xen-builder.localnet:/usr/home/julian/src/obj/usr/home/julian/sr > c/head/sys/BLITZXEN > WARNING: WITNESS option enabled, expect reduced performance. > Xen reported: 701.590 MHz processor. > Timecounter "ixen" frequency 10 Hz quality 0 > CPU: Intel Pentium III Xeon (701.59-MHz 686-class CPU) > Origin = "GenuineIntel" Id = 0x6a1 Stepping = 1 > > Features=0x383fbff > real memory = 268435456 (256 MB) > avail memory = 254689280 (242 MB) > cpu=0 irq=0 vector=0 > cpu=0 irq=0 vector=1 > kbd0 at kbdmux0 > xenbus0: on motherboard > xc0: on motherboard > Timecounters tick every 100.000 msec > xbd0: 4096MB at device/vbd/51713 on xenbus0 > xn0: at device/vif/0 on xenbus0 > xn0: Ethernet address: aa:00:4b:17:23:8e > [XEN] netfront_backend_changed: newstate=2 > > Btw, it is very cool to see FreeBSD on Xen. :) > > Regards, > -- > Julian Stecklina > > Well, take it from an old hand: the only reason it would be easier to > program in C is that you can't easily express complex problems in C, > so you don't. - Erik Naggum (in comp.lang.lisp) > > ___ > freebsd-xen@freebsd.org mailing list > http://lists.freebsd.org/mailman/listinfo/freebsd-xen > To unsubscribe, send any mail to "freebsd-xen-unsubscr...@freebsd.org" > ___ freebsd-xen@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-xen To unsubscribe, send any mail to "freebsd-xen-unsubscr...@freebsd.org"
Re: kern.hz = 10
Mister Olli writes: > Hi, > > I've never experimented with the kern.hz variable, but on all > configuration examples for 8-CURRENT if always seen 'kern.hz=100'. > >>From where did you get your configuration? I use the default config (albeit renamed to BLITZXEN) from 8-CURRENT. Regards, -- Julian Stecklina Well, take it from an old hand: the only reason it would be easier to program in C is that you can't easily express complex problems in C, so you don't. - Erik Naggum (in comp.lang.lisp) ___ freebsd-xen@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-xen To unsubscribe, send any mail to "freebsd-xen-unsubscr...@freebsd.org"
Re: kern.hz = 10
Hi, I've never experimented with the kern.hz variable, but on all configuration examples for 8-CURRENT if always seen 'kern.hz=100'. >From where did you get your configuration? -- Mr. Olli Am Freitag, den 30.01.2009, 02:08 +0100 schrieb Julian Stecklina: > Hello, > > I have an 8-CURRENT (as of yesterday) DomU running. kern.hz is set to > 10. Is this intended? This gives some strange effects: > > $ ping www.google.de > PING www.l.google.com (74.125.39.99): 56 data bytes > 64 bytes from 74.125.39.99: icmp_seq=0 ttl=246 time=100.000 ms > 64 bytes from 74.125.39.99: icmp_seq=1 ttl=246 time=0.000 ms > 64 bytes from 74.125.39.99: icmp_seq=2 ttl=246 time=0.000 ms > > dmesg: > FreeBSD is a registered trademark of The FreeBSD Foundation. > FreeBSD 8.0-CURRENT #0 r187877: Thu Jan 29 16:32:39 CET 2009 > > jul...@bsd-xen-builder.localnet:/usr/home/julian/src/obj/usr/home/julian/sr > c/head/sys/BLITZXEN > WARNING: WITNESS option enabled, expect reduced performance. > Xen reported: 701.590 MHz processor. > Timecounter "ixen" frequency 10 Hz quality 0 > CPU: Intel Pentium III Xeon (701.59-MHz 686-class CPU) > Origin = "GenuineIntel" Id = 0x6a1 Stepping = 1 > > Features=0x383fbff > real memory = 268435456 (256 MB) > avail memory = 254689280 (242 MB) > cpu=0 irq=0 vector=0 > cpu=0 irq=0 vector=1 > kbd0 at kbdmux0 > xenbus0: on motherboard > xc0: on motherboard > Timecounters tick every 100.000 msec > xbd0: 4096MB at device/vbd/51713 on xenbus0 > xn0: at device/vif/0 on xenbus0 > xn0: Ethernet address: aa:00:4b:17:23:8e > [XEN] netfront_backend_changed: newstate=2 > > Btw, it is very cool to see FreeBSD on Xen. :) > > Regards ___ freebsd-xen@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-xen To unsubscribe, send any mail to "freebsd-xen-unsubscr...@freebsd.org"
kern.hz = 10
Hello, I have an 8-CURRENT (as of yesterday) DomU running. kern.hz is set to 10. Is this intended? This gives some strange effects: $ ping www.google.de PING www.l.google.com (74.125.39.99): 56 data bytes 64 bytes from 74.125.39.99: icmp_seq=0 ttl=246 time=100.000 ms 64 bytes from 74.125.39.99: icmp_seq=1 ttl=246 time=0.000 ms 64 bytes from 74.125.39.99: icmp_seq=2 ttl=246 time=0.000 ms dmesg: FreeBSD is a registered trademark of The FreeBSD Foundation. FreeBSD 8.0-CURRENT #0 r187877: Thu Jan 29 16:32:39 CET 2009 jul...@bsd-xen-builder.localnet:/usr/home/julian/src/obj/usr/home/julian/sr c/head/sys/BLITZXEN WARNING: WITNESS option enabled, expect reduced performance. Xen reported: 701.590 MHz processor. Timecounter "ixen" frequency 10 Hz quality 0 CPU: Intel Pentium III Xeon (701.59-MHz 686-class CPU) Origin = "GenuineIntel" Id = 0x6a1 Stepping = 1 Features=0x383fbff real memory = 268435456 (256 MB) avail memory = 254689280 (242 MB) cpu=0 irq=0 vector=0 cpu=0 irq=0 vector=1 kbd0 at kbdmux0 xenbus0: on motherboard xc0: on motherboard Timecounters tick every 100.000 msec xbd0: 4096MB at device/vbd/51713 on xenbus0 xn0: at device/vif/0 on xenbus0 xn0: Ethernet address: aa:00:4b:17:23:8e [XEN] netfront_backend_changed: newstate=2 Btw, it is very cool to see FreeBSD on Xen. :) Regards, -- Julian Stecklina Well, take it from an old hand: the only reason it would be easier to program in C is that you can't easily express complex problems in C, so you don't. - Erik Naggum (in comp.lang.lisp) ___ freebsd-xen@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-xen To unsubscribe, send any mail to "freebsd-xen-unsubscr...@freebsd.org"