Re: smp in 5.1
At 6:16 PM -0400 8/11/03, Eriq Lamar wrote: Is there any advantage in 5.1 over 4.8 for two amd mp's. and if so could someone tell what they are. I am interested in building dual system using mp's but not sure which version would be better. I run 5.x on a dual-Althon 2000 machine. I have no idea if that gives me much of a benefit as far as *smp* is concerned, but it certainly works fine. And there are other aspects to 5.x which are very attractive to me, such as no more MAKEDEV, and the inclusion of filesystem-snapshots. -- Garance Alistair Drosehn= [EMAIL PROTECTED] Senior Systems Programmer or [EMAIL PROTECTED] Rensselaer Polytechnic Instituteor [EMAIL PROTECTED] ___ [EMAIL PROTECTED] mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to [EMAIL PROTECTED]
Re: smp in 5.1
On Tue, Aug 12, 2003 at 08:25:38AM +1000, Andy Farkas wrote: On Mon, 11 Aug 2003, Eriq Lamar wrote: Is there any advantage in 5.1 over 4.8 for two amd mp's. and if so could someone tell what they are. I am interested in building dual system using mp's but not sure which version would be better. Scheduling in 5.1 is broken (sched_ule doesn't even work*). Stick with 4.8. * for me, sched_ule completely locks up my box, no ping, no keybd. Exact same kernel with sched_4bsd works fine. Erm, this in itself isn't a reason to avoid 5.1, since as you noted sched_4bsd works fine. Kris pgp0.pgp Description: PGP signature
Re: smp in 5.1
On Tue, Aug 12, 2003 at 08:25:38AM +1000, Andy Farkas wrote: On Mon, 11 Aug 2003, Eriq Lamar wrote: Is there any advantage in 5.1 over 4.8 for two amd mp's. and if so could someone tell what they are. I am interested in building dual system using mp's but not sure which version would be better. Scheduling in 5.1 is broken (sched_ule doesn't even work*). Stick with 4.8. * for me, sched_ule completely locks up my box, no ping, no keybd. Exact same kernel with sched_4bsd works fine. Strange. ULE has worked fine on my UP system for several months and the SMP system I recently obtained from a co-worker hasn't panicked while running ULE. Can you drop into ddb and trace the problem with ULE on your system? -- Steve ___ [EMAIL PROTECTED] mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to [EMAIL PROTECTED]
RE: smp in 5.1
I've got a dual athlon 1900+ MP. I've run both 4.x and 5.x. On 5.x I've tried both schedulers (4BSD and ULE). I don't think I've ever had any crashes, except when I install nvidia-driver, and then it crashes all the time. Theoretically 5.x should be better for smp as much work has happened in terms of locking. I haven't done any benchmarks, but I wouldn't expect any dramatic improvement, as locking work in many subsystems is still incomplete. Many things still need GIANT. 5.x does have a bunch of other good stuff though, and going that route saves you from upgrading later through an even bigger version gap. That's my two sense, and if it sounds like I know what I'm talking about, it's pure coincidence ;-) Others: please correct me if I got anything really wrong (I tried to be vague enough to aviod that, but we'll see). ;-) Evan Dower From: Andre Guibert de Bruet [EMAIL PROTECTED] To: [EMAIL PROTECTED] Subject: RE: smp in 5.1 Date: Mon, 11 Aug 2003 22:40:21 -0400 (EDT) I guess I'll chime in as well... I have a Dual Athlon 2000+ MP here and it's running like a charm with SCHED_4BSD. Andy Andre Guibert de Bruet | Enterprise Software Consultant Silicon Landmark, LLC. | http://siliconlandmark.com/ On Mon, 11 Aug 2003, derwood wrote: I've been running 5.1-Current since its release on a Dell Precision 410 with dual P-III 500's No SMP problems here at all.. Its been extremely stable for me thus far. Darin - -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Steve Kargl Sent: Monday, August 11, 2003 7:10 PM To: Andy Farkas Cc: [EMAIL PROTECTED]; Eriq Lamar Subject: Re: smp in 5.1 On Tue, Aug 12, 2003 at 08:25:38AM +1000, Andy Farkas wrote: On Mon, 11 Aug 2003, Eriq Lamar wrote: Is there any advantage in 5.1 over 4.8 for two amd mp's. and if so could someone tell what they are. I am interested in building dual system using mp's but not sure which version would be better. Scheduling in 5.1 is broken (sched_ule doesn't even work*). Stick with 4.8. * for me, sched_ule completely locks up my box, no ping, no keybd. Exact same kernel with sched_4bsd works fine. Strange. ULE has worked fine on my UP system for several months and the SMP system I recently obtained from a co-worker hasn't panicked while running ULE. Can you drop into ddb and trace the problem with ULE on your system? -- Steve ___ [EMAIL PROTECTED] mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to [EMAIL PROTECTED] ___ [EMAIL PROTECTED] mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to [EMAIL PROTECTED] ___ [EMAIL PROTECTED] mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to [EMAIL PROTECTED] _ Tired of spam? Get advanced junk mail protection with MSN 8. http://join.msn.com/?page=features/junkmail ___ [EMAIL PROTECTED] mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to [EMAIL PROTECTED]
RE: smp in 5.1
I guess I'll chime in as well... I have a Dual Athlon 2000+ MP here and it's running like a charm with SCHED_4BSD. Andy Andre Guibert de Bruet | Enterprise Software Consultant Silicon Landmark, LLC. | http://siliconlandmark.com/ On Mon, 11 Aug 2003, derwood wrote: I've been running 5.1-Current since its release on a Dell Precision 410 with dual P-III 500's No SMP problems here at all.. Its been extremely stable for me thus far. Darin - -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Steve Kargl Sent: Monday, August 11, 2003 7:10 PM To: Andy Farkas Cc: [EMAIL PROTECTED]; Eriq Lamar Subject: Re: smp in 5.1 On Tue, Aug 12, 2003 at 08:25:38AM +1000, Andy Farkas wrote: On Mon, 11 Aug 2003, Eriq Lamar wrote: Is there any advantage in 5.1 over 4.8 for two amd mp's. and if so could someone tell what they are. I am interested in building dual system using mp's but not sure which version would be better. Scheduling in 5.1 is broken (sched_ule doesn't even work*). Stick with 4.8. * for me, sched_ule completely locks up my box, no ping, no keybd. Exact same kernel with sched_4bsd works fine. Strange. ULE has worked fine on my UP system for several months and the SMP system I recently obtained from a co-worker hasn't panicked while running ULE. Can you drop into ddb and trace the problem with ULE on your system? -- Steve ___ [EMAIL PROTECTED] mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to [EMAIL PROTECTED] ___ [EMAIL PROTECTED] mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to [EMAIL PROTECTED] ___ [EMAIL PROTECTED] mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to [EMAIL PROTECTED]
Re: smp in 5.1
On Mon, 11 Aug 2003, Eriq Lamar wrote:
Is there any advantage in 5.1 over 4.8 for two amd mp's. and if so could
someone tell what they are. I am interested in building dual system using
mp's but not sure which version would be better.
Scheduling in 5.1 is broken (sched_ule doesn't even work*).
Stick with 4.8.
* for me, sched_ule completely locks up my box, no ping, no keybd. Exact
same kernel with sched_4bsd works fine.
--
:{ [EMAIL PROTECTED]
Andy Farkas
System Administrator
Speednet Communications
http://www.speednet.com.au/
___
[EMAIL PROTECTED] mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-current
To unsubscribe, send any mail to [EMAIL PROTECTED]
RE: smp in 5.1
I've been running 5.1-Current since its release on a Dell Precision 410 with dual P-III 500's No SMP problems here at all.. Its been extremely stable for me thus far. Darin - -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Steve Kargl Sent: Monday, August 11, 2003 7:10 PM To: Andy Farkas Cc: [EMAIL PROTECTED]; Eriq Lamar Subject: Re: smp in 5.1 On Tue, Aug 12, 2003 at 08:25:38AM +1000, Andy Farkas wrote: On Mon, 11 Aug 2003, Eriq Lamar wrote: Is there any advantage in 5.1 over 4.8 for two amd mp's. and if so could someone tell what they are. I am interested in building dual system using mp's but not sure which version would be better. Scheduling in 5.1 is broken (sched_ule doesn't even work*). Stick with 4.8. * for me, sched_ule completely locks up my box, no ping, no keybd. Exact same kernel with sched_4bsd works fine. Strange. ULE has worked fine on my UP system for several months and the SMP system I recently obtained from a co-worker hasn't panicked while running ULE. Can you drop into ddb and trace the problem with ULE on your system? -- Steve ___ [EMAIL PROTECTED] mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to [EMAIL PROTECTED] ___ [EMAIL PROTECTED] mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to [EMAIL PROTECTED]
Re: SMP in 5.1 cant deactivate hyperthreading
On Thu, 12 Jun 2003, Killing wrote: Halting them will still cause the CPUs to be detected. They just won't do any useful work. Yep but the issue is that all the core admin tools are unaware of this and hence include the virtual cores in idle calcs etc making load monitoring impossible without nasty cludges :( So what's the way forward? 1. Dont just use halt have a compile or other directive to disable them? 2. Update all tools to be halt aware? Personally I'd go with 2 all be it more work / ramifications on other 3rd party tools as it gives the benefit of also working when physical CPU's are halted. Guh. Checking with some people here, it appears that the HT stuff is still undergoing some tweaking. I'd suggest that you stay tuned for any commits from john baldwin re: HT. Your case appears to be unique. -- Doug White| FreeBSD: The Power to Serve [EMAIL PROTECTED] | www.FreeBSD.org ___ [EMAIL PROTECTED] mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to [EMAIL PROTECTED]
Re: SMP in 5.1 cant deactivate hyperthreading
On Wed, 11 Jun 2003, Steven Hartland wrote: sysctl machdep.hlt_logical_cpus: machdep.hlt_logical_cpus: 1 Halting them will still cause the CPUs to be detected. They just won't do any useful work. Relevant sections from dmesg: Programming 24 pins in IOAPIC #0 IOAPIC #0 intpin 2 - irq 0 Programming 24 pins in IOAPIC #1 Programming 24 pins in IOAPIC #2 FreeBSD/SMP: Multiprocessor System Detected: 4 CPUs cpu0 (BSP): apic id: 0, version: 0x00050014, at 0xfee0 cpu1 (AP): apic id: 6, version: 0x00050014, at 0xfee0 cpu2 (AP): apic id: 1, version: 0x00050014, at 0xfee0 cpu3 (AP): apic id: 7, version: 0x00050014, at 0xfee0 -- Doug White| FreeBSD: The Power to Serve [EMAIL PROTECTED] | www.FreeBSD.org ___ [EMAIL PROTECTED] mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to [EMAIL PROTECTED]
Re: SMP in 5.1 cant deactivate hyperthreading
- Original Message - From: Doug White [EMAIL PROTECTED] sysctl machdep.hlt_logical_cpus: machdep.hlt_logical_cpus: 1 Halting them will still cause the CPUs to be detected. They just won't do any useful work. Yep but the issue is that all the core admin tools are unaware of this and hence include the virtual cores in idle calcs etc making load monitoring impossible without nasty cludges :( So what's the way forward? 1. Dont just use halt have a compile or other directive to disable them? 2. Update all tools to be halt aware? Personally I'd go with 2 all be it more work / ramifications on other 3rd party tools as it gives the benefit of also working when physical CPU's are halted. Which ever it needs someone to pick it up ASAP dont you think? Steve ___ [EMAIL PROTECTED] mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to [EMAIL PROTECTED]
Re: SMP in 5.1 cant deactivate hyperthreading
On Thu, 12 Jun 2003, Killing wrote: - Original Message - From: Doug White [EMAIL PROTECTED] sysctl machdep.hlt_logical_cpus: machdep.hlt_logical_cpus: 1 Halting them will still cause the CPUs to be detected. They just won't do any useful work. Yep but the issue is that all the core admin tools are unaware of this and hence include the virtual cores in idle calcs etc making load monitoring impossible without nasty cludges :( So what's the way forward? 1. Dont just use halt have a compile or other directive to disable them? 2. Update all tools to be halt aware? Personally I'd go with 2 all be it more work / ramifications on other 3rd party tools as it gives the benefit of also working when physical CPU's are halted. Well, hyperthreading can be disabled via a kernel directive, right? Which ever it needs someone to pick it up ASAP dont you think? Really? What's the hurry? FreeBSD 5.x isn't even bootable/installable on a number of SMP machines yet (ex. Dell Poweredge 6350). Steve Tom ___ [EMAIL PROTECTED] mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to [EMAIL PROTECTED]
Re: SMP in 5.1 cant deactivate hyperthreading
Well, hyperthreading can be disabled via a kernel directive, right? From what I've seen that was removed between 5.0 and 5.1 correct me if Im wrong. Which ever it needs someone to pick it up ASAP dont you think? Really? What's the hurry? FreeBSD 5.x isn't even bootable/installable on a number of SMP machines yet (ex. Dell Poweredge 6350). I suppose the hurry is that basic utils that we use day to day like top and vmstat to monitor machine load cannot be trusted to give accurate info. Yes 5.X is still new tech and may not run on all machines but on the ones which it does ( and it runs very well here ) basic tools are required. If it doesn't run on a machine your under know false impressions, if however you users complain of performance issues and you look @ top and it says 50% idle and its really 0% idle its a different matter. Steve / K ___ [EMAIL PROTECTED] mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to [EMAIL PROTECTED]
Re: SMP in 5.1 cant deactivate hyperthreading
Killing wrote this message on Thu, Jun 12, 2003 at 18:47 +0100: I suppose the hurry is that basic utils that we use day to day like top and vmstat to monitor machine load cannot be trusted to give accurate info. Actually, the basic tools ARE correct, there is a cpu sitting idle that the sysadmin disabled. So, if the sysadmin would reenable the cpu, then jobs could be dispatached to it. Would you rather some junior admin go and disable the cpu, and then six months later wondering why the performance is so slow? Yes 5.X is still new tech and may not run on all machines but on the ones which it does ( and it runs very well here ) basic tools are required. If it doesn't run on a machine your under know false impressions, if however you users complain of performance issues and you look @ top and it says 50% idle and its really 0% idle its a different matter. That's a good way to remind the admin to turn the cpu back on. Now is there any good reason why you need to keep the cpu disabled? -- John-Mark Gurney Voice: +1 415 225 5579 All that I will do, has been done, All that I have, has not. ___ [EMAIL PROTECTED] mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to [EMAIL PROTECTED]
Re: SMP in 5.1 cant deactivate hyperthreading
Original Message - From: John-Mark Gurney [EMAIL PROTECTED] Yes 5.X is still new tech and may not run on all machines but on the ones which it does ( and it runs very well here ) basic tools are required. If it doesn't run on a machine your under know false impressions, if however you users complain of performance issues and you look @ top and it says 50% idle and its really 0% idle its a different matter. That's a good way to remind the admin to turn the cpu back on. Don't by it there just checking sysctl is easy enough. Now is there any good reason why you need to keep the cpu disabled? Its disabled by default due to performance reasons and we keep it that way. We are talking hyperthreading, logical CPU's, not real physical CPU's. Its quite easy to see if CPU's are disabled as mentioned above. If you check the release notes for top then you'll see a similar fix had to be made for sunos5 in beta5 iirc. Steve ___ [EMAIL PROTECTED] mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to [EMAIL PROTECTED]
