|
Hello,
I've got a new Supermicro X58 system with an Intel Core i7 930 with 6 GB ram that is not performing nearly as fast as it should in many ways (compiling, network transfers). To give an example, it has been building the gcc44 port for about 10 hours now and at the same time rsync'ing from a Linux box on the same Gigabit network is only getting throughput of between 10-25 MB/sec. When I did a buildkernel for 8-STABLE, it took 17 hours! In the BIOS, I have played with a few settings and some actually made it worse. What I have done now is disabled Hyperthreading and Speedstep. I have attached the dmesg details, and here's some system info and then some system stats: bryce@tahiti[~]>uname -a FreeBSD tahiti.bryce.net 8.0-STABLE FreeBSD 8.0-STABLE #0: Wed Apr 28 10:53:37 CDT 2010 [hidden email]:/usr/obj/usr/src/sys/GENERIC amd64 bryce@tahiti[~]>cat /boot/loader.conf ahci_load="YES" ichsmb_load="YES" smb_load="YES" coretemp_load="YES" zfs_load="YES" vfs.root.mountfrom="zfs:system" hint.p4tcc.0.disabled=1 hint.acpi_throttle.0.disabled=1 bryce@tahiti[~]>cat /etc/sysctl.conf kern.timecounter.hardware=HPET bryce@tahiti[~]>vmstat 1 procs memory page disks faults cpu r b w avm fre flt re pi po fr sr ad0 ad1 in sy cs us sy id 5 0 0 1068M 3478M 572 1 1 0 862 0 0 0 9370 16514 16157 71 22 7 5 0 0 1068M 3478M 2 0 0 0 0 0 0 0 8008 14504 11716 81 17 2 5 0 0 1068M 3478M 0 0 0 0 0 0 0 0 12429 22323 18125 77 23 0 5 0 0 1068M 3478M 0 0 0 0 0 0 0 0 12348 22125 17988 73 27 0 bryce@tahiti[~]>vmstat -i interrupt total rate irq1: atkbd0 9291 0 irq17: fwohci0 1 0 cpu0: timer 75416246 2000 irq256: em0 137590284 3649 irq257: em0 206367605 5473 irq260: em0 1 0 irq266: ahci0 9892384 262 cpu2: timer 75415653 2000 cpu3: timer 75415702 2000 cpu1: timer 75415561 2000 Total 655522728 17385 bryce@tahiti[~]>netstat -I em0 -h 1 input (em0) output packets errs idrops bytes packets errs bytes colls 7.7K 0 0 11M 7.2K 0 475K 0 8.1K 0 0 12M 7.4K 0 491K 0 7.8K 0 0 11M 7.2K 0 476K 0 bryce@tahiti[/usr/adm]>iostat 1 tty ada0 ada1 ada2 cpu tin tout KB/t tps MB/s KB/t tps MB/s KB/t tps MB/s us ni sy in id 0 108 22.35 3 0.07 20.61 3 0.07 58.60 0 0.00 71 0 4 17 7 0 222 64.00 1 0.06 128.00 1 0.12 0.00 0 0.00 87 0 2 11 0 _______________________________________________ [hidden email] mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-performance To unsubscribe, send any mail to "[hidden email]" |
|
On May 1, 2010, at 11:10 , Bryce Edwards wrote: > Hello, > > I've got a new Supermicro X58 system with an Intel Core i7 930 with 6 > GB ram that is not performing nearly as fast as it should in many ways > (compiling, network transfers). To give an example, it has been > building the gcc44 port for about 10 hours now and at the same time > rsync'ing from a Linux box on the same Gigabit network is only getting > throughput of between 10-25 MB/sec. When I did a buildkernel for > 8-STABLE, it took 17 hours! > > In the BIOS, I have played with a few settings and some actually made > it worse. What I have done now is disabled Hyperthreading and > Speedstep. I have attached the dmesg details, and here's some system > info and then some system stats: > > > bryce@tahiti[~]>uname -a > FreeBSD tahiti.bryce.net 8.0-STABLE FreeBSD 8.0-STABLE #0: Wed Apr 28 > 10:53:37 CDT 2010 > [hidden email]:/usr/obj/usr/src/sys/GENERIC amd64 > > > > bryce@tahiti[~]>cat /boot/loader.conf > ahci_load="YES" > ichsmb_load="YES" > smb_load="YES" > coretemp_load="YES" > > zfs_load="YES" > vfs.root.mountfrom="zfs:system" > > hint.p4tcc.0.disabled=1 > hint.acpi_throttle.0.disabled=1 > > > bryce@tahiti[~]>cat /etc/sysctl.conf > kern.timecounter.hardware=HPET > > > bryce@tahiti[~]>vmstat 1 > procs memory page disks faults cpu > r b w avm fre flt re pi po fr sr ad0 ad1 in sy > cs us sy id > 5 0 0 1068M 3478M 572 1 1 0 862 0 0 0 9370 16514 > 16157 71 22 7 > 5 0 0 1068M 3478M 2 0 0 0 0 0 0 0 8008 14504 > 11716 81 17 2 > 5 0 0 1068M 3478M 0 0 0 0 0 0 0 0 12429 22323 > 18125 77 23 0 > 5 0 0 1068M 3478M 0 0 0 0 0 0 0 0 12348 22125 > 17988 73 27 0 > > > bryce@tahiti[~]>vmstat -i > interrupt total rate > irq1: atkbd0 9291 0 > irq17: fwohci0 1 0 > cpu0: timer 75416246 2000 > irq256: em0 137590284 3649 > irq257: em0 206367605 5473 > irq260: em0 1 0 > irq266: ahci0 9892384 262 > cpu2: timer 75415653 2000 > cpu3: timer 75415702 2000 > cpu1: timer 75415561 2000 > Total 655522728 17385 > > > bryce@tahiti[~]>netstat -I em0 -h 1 > input (em0) output > packets errs idrops bytes packets errs bytes colls > 7.7K 0 0 11M 7.2K 0 475K 0 > 8.1K 0 0 12M 7.4K 0 491K 0 > 7.8K 0 0 11M 7.2K 0 476K 0 > > bryce@tahiti[/usr/adm]>iostat 1 > tty ada0 ada1 ada2 cpu > tin tout KB/t tps MB/s KB/t tps MB/s KB/t tps MB/s us ni sy in id > 0 108 22.35 3 0.07 20.61 3 0.07 58.60 0 0.00 71 0 4 17 7 > 0 222 64.00 1 0.06 128.00 1 0.12 0.00 0 0.00 87 0 2 11 0 First thing to try. Turn off Legacy USB in the BIOS. Is this a 2 CPU machine with only one CPU installed? Make sure your memory is all in the correct slots, on modern (non Front Side Bus) machines like Nehalem only the memory in the slots that are adjacent to a CPU are seen by the CPU. I gather you have a kernel with INVARIANTS, WITNESS and the like turned off, correct? Best, George _______________________________________________ [hidden email] mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-performance To unsubscribe, send any mail to "[hidden email]" |
|
In reply to this post by Bryce Edwards
I was a little surprised to read this, I've been running 7.1 based
code on some supermicro nehalem machines (5500, Xeon, X8DT6/X8DTE) which perform reasonably well, and was hoping to test some X8SAX based systems with core i7 to compare performance. The configuration Ive been using so far: 2x E5520 Xeon, SMT disabled ie. 2 x 4 cores 24GB (3x 4 GB per CPU socket) LSILOGIC SASX36 A1 7015, SEAGATE ST9300603SS 0006 This made me curious enough to put 8.0-RELEASE on one of these to find out how it would do on buildkernel, and it seems to perform 'as expected', nhm005# /usr/bin/time -h make buildkernel 18m20.31s real 13m13.84s user 5m1.47s sys nhm005# /usr/bin/time -h make -j4 buildkernel 8m53.96s real 13m22.34s user 6m54.47s sys A misbehaving mpt controller/driver seemed to slow this down a bit, it was logging full queue events during the initial install and continued to do so during buildkernel. mpt0: QUEUE FULL EVENT: Bus 0x00 Target 0x00 Depth 127 mpt0: QUEUE FULL EVENT: Bus 0x00 Target 0x00 Depth 128 mpt0: QUEUE FULL EVENT: Bus 0x00 Target 0x00 Depth 128 I think it's safe to say the x8sax tests just moved up on my TODO list a little, unfortunately that hardware is tied up at the moment. I'd be interested to see lmbench results if you have any you want to share. (/usr/ports/benchmarks/lmbench) I'll attach dmesg & lmbench results for the X8DT6/E5520 configuration. Case _______________________________________________ [hidden email] mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-performance To unsubscribe, send any mail to "[hidden email]" |
|
In reply to this post by George Neville-Neil
Bingo, looks it was the Legacy USB setting in the BIOS. I had it on
"Auto" (default) and the docs say it will disable unless there is legacy USB devices. I have now put all the other BIOS options to default (except enabling AHCI) and here's what buildworld now looks like: 2h39m29.20s real 11h35m35.36s user 3h28m19.15s sys Now I need to go research more about this to understand more about the Legacy USB behavior. ::Bryce:: On Mon, May 3, 2010 at 1:34 PM, George Neville-Neil <[hidden email]> wrote: > > On May 1, 2010, at 11:10 , Bryce Edwards wrote: > >> Hello, >> >> I've got a new Supermicro X58 system with an Intel Core i7 930 with 6 >> GB ram that is not performing nearly as fast as it should in many ways >> (compiling, network transfers). To give an example, it has been >> building the gcc44 port for about 10 hours now and at the same time >> rsync'ing from a Linux box on the same Gigabit network is only getting >> throughput of between 10-25 MB/sec. When I did a buildkernel for >> 8-STABLE, it took 17 hours! >> >> In the BIOS, I have played with a few settings and some actually made >> it worse. What I have done now is disabled Hyperthreading and >> Speedstep. I have attached the dmesg details, and here's some system >> info and then some system stats: >> >> >> bryce@tahiti[~]>uname -a >> FreeBSD tahiti.bryce.net 8.0-STABLE FreeBSD 8.0-STABLE #0: Wed Apr 28 >> 10:53:37 CDT 2010 >> [hidden email]:/usr/obj/usr/src/sys/GENERIC amd64 >> >> >> >> bryce@tahiti[~]>cat /boot/loader.conf >> ahci_load="YES" >> ichsmb_load="YES" >> smb_load="YES" >> coretemp_load="YES" >> >> zfs_load="YES" >> vfs.root.mountfrom="zfs:system" >> >> hint.p4tcc.0.disabled=1 >> hint.acpi_throttle.0.disabled=1 >> >> >> bryce@tahiti[~]>cat /etc/sysctl.conf >> kern.timecounter.hardware=HPET >> >> >> bryce@tahiti[~]>vmstat 1 >> procs memory page disks faults cpu >> r b w avm fre flt re pi po fr sr ad0 ad1 in sy >> cs us sy id >> 5 0 0 1068M 3478M 572 1 1 0 862 0 0 0 9370 16514 >> 16157 71 22 7 >> 5 0 0 1068M 3478M 2 0 0 0 0 0 0 0 8008 14504 >> 11716 81 17 2 >> 5 0 0 1068M 3478M 0 0 0 0 0 0 0 0 12429 22323 >> 18125 77 23 0 >> 5 0 0 1068M 3478M 0 0 0 0 0 0 0 0 12348 22125 >> 17988 73 27 0 >> >> >> bryce@tahiti[~]>vmstat -i >> interrupt total rate >> irq1: atkbd0 9291 0 >> irq17: fwohci0 1 0 >> cpu0: timer 75416246 2000 >> irq256: em0 137590284 3649 >> irq257: em0 206367605 5473 >> irq260: em0 1 0 >> irq266: ahci0 9892384 262 >> cpu2: timer 75415653 2000 >> cpu3: timer 75415702 2000 >> cpu1: timer 75415561 2000 >> Total 655522728 17385 >> >> >> bryce@tahiti[~]>netstat -I em0 -h 1 >> input (em0) output >> packets errs idrops bytes packets errs bytes colls >> 7.7K 0 0 11M 7.2K 0 475K 0 >> 8.1K 0 0 12M 7.4K 0 491K 0 >> 7.8K 0 0 11M 7.2K 0 476K 0 >> >> bryce@tahiti[/usr/adm]>iostat 1 >> tty ada0 ada1 ada2 cpu >> tin tout KB/t tps MB/s KB/t tps MB/s KB/t tps MB/s us ni sy in id >> 0 108 22.35 3 0.07 20.61 3 0.07 58.60 0 0.00 71 0 4 17 7 >> 0 222 64.00 1 0.06 128.00 1 0.12 0.00 0 0.00 87 0 2 11 0 > > First thing to try. Turn off Legacy USB in the BIOS. > > Is this a 2 CPU machine with only one CPU installed? Make sure your memory is all > in the correct slots, on modern (non Front Side Bus) machines like Nehalem > only the memory in the slots that are adjacent to a CPU are seen by the CPU. > > I gather you have a kernel with INVARIANTS, WITNESS and the like turned off, correct? > > Best, > George > > [hidden email] mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-performance To unsubscribe, send any mail to "[hidden email]" |
| Powered by Nabble | Edit this page |
