Re: FreeBSD 6.0, amd64, A8N-SLI and 4gb ram
Joseph Kerian wrote: I was planning on purchasing one of these boards in the near future, so I'm rather interested if you have solved this particular problem. In no particular order, my suggestions are: -The A8N-SLI's are extremely picky about the RAM you give them; have you double checked with a memtest run to verify that it's not the problem? No, but I think it is purely down to the bios to enable the last 1gb or not. If I disable both the h/w and s/w PAE options in the bios then the bios only reports 3gb available and FreeBSD the same and it runs ok (sans the memory leak problem I've also emailed the question list about - with little response). If I enable *only* the h/w PAE option in the bios, both it and FreeBSD report 4gb available, but FreeBSD is visibly slower on startup, the nic doesn't appear to work correctly (DHCP fails) and it is very flaky after login. If I enable *only* the s/w PA option in the bios, both it and FreeBSD report 4gb available, but FreeBSD fails to complete startup, spewing 'entry of nVidia Mediashield metadata is NOT supported' on the console until manually rebooted. Also, what happens if the new RAM is the only RAM in the system? Everything works as expected (2gb available/used), no problems encountered. -Does anything change if you compile for a single processor? FreeBSD should run fine even if it's not taking advantage of the dual cored nature of the processor. Haven't tried that, but I have the original kernel on the box, so I will. -There appears to have been a bios problem with using exactly 4 gigs of ram in the A8N Deluxe boards at one point, not sure if this also showed up on the Premiums or not. You may want to verify that you are running the most recent bios. Yep, I'm running 1009 which is the latest non-beta version. I found the cause of the invalid option error. From the handbook section on the PAE option: *Note:* The PAE support in FreeBSD is only available for Intel IA-32 processors. Note that it appears that PAE is a system for addressing _more_ than 4 gigs of RAM (handbook section 8). Since your board only supports up to 4 gigs ( http://www.asus.com/products4.aspx?modelmenu=2model=539l1=3l2=15l3=148), I guess I'm curious why the PAE options exist in the bios. The bios help(!) says '4gb or more' and it appears to be the only way to get it to enable the last 1gb. After some discussion on the Ubuntu 64 forum I'm going to try the Live CD 64 bit to see if it boots up ok with the 4gb turned on in the bios and if so, it looks like I will have to move to that distro to get this box working properly. Robert ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to [EMAIL PROTECTED]
Re: Memory leak?
On Mon, 13 Feb 2006 09:29:03 +0100, Erik Norgaard [EMAIL PROTECTED] said: Do you run other applications also? There was a discussion on CURRENT@ some weeks ago about a memory leak that turned out to be firefox with some extensions, updates are available now. Unfortunately no, its cli only, no x, pretty much just Postgres and Python and C :-( Robert ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to [EMAIL PROTECTED]
Re: Memory leak?
On Mon, 13 Feb 2006 16:40:54 -0500, David Scheidt [EMAIL PROTECTED] said: I've seen (very, very, very, very) large memory leaks on long-lived Python processes. I haven't looked at it to figure out if it's python, some module, or the application doing something stupid. But the processes will grow until they hit their limits. What's your definition of long-lived? My scenario is that I'm processing a particular dataset in Python which is launched by a shell script, once finished (after 30-35mins) the Python app completes and the shell script launches another instance on a new dataset. All memory allocated by the finished Python app should be freed/made inactive shouldn't it? Here's some more data: After a reboot this is what top says: Mem: 45M Active, 13M Inact, 61M Wired, 4K Cache, 60M Buf, 2842M Free Swap: 4068M Total, 4068M Free which totals 3021M After 1 dataset it is: Mem: 107M Active, 1919M Inact, 158M Wired, 16K Cache, 214M Buf, 570M Free Swap: 4068M Total, 4068M Free which totals 2968M While running on the 6th dataset: Mem: 1032M Active, 1045M Inact, 260M Wired, 145M Cache, 214M Buf, 4664K Free Swap: 4068M Total, 108K Used, 4068M Free which totals 2700.6M Are my assumptions incorrect, should the totals displayed by top be at least approximately equal? Robert ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to [EMAIL PROTECTED]
Re: Memory leak?
On Mon, 13 Feb 2006 17:58:07 -0500, Charles Swiger [EMAIL PROTECTED] said: On Feb 13, 2006, at 5:13 PM, Robert Leftwich wrote: Possibly your database is using lots of SysV shared memory, which would explain why wired is going up so much, otherwise perhaps something in the kernel is leaking. sysctl kern.malloc might be interesting to consider. What should I be looking for? The maximum MemUse is devbuf 2039 8340K, the max InUse is sysctloid 3613 176K. Are my assumptions incorrect, should the totals displayed by top be at least approximately equal? Exclude the buf entry from your math, that will be closer. You should be looking further down at the SIZE column to see which processes are using so much RAM... I can't see anything that explains the discrepancy. Below is the top -o size after a reboot, followed by the current top after 8 datasets (the extra python process is the analysis app - at a low memory usage point): (Note that the original 2 python processes are web servers and that I have 3 postgres clusters running on different ports, pending a move to separate machines - assuming I can solve this problem) PID USERNAME THR PRI NICE SIZERES STATE C TIME WCPU COMMAND 599 msf 1 40 75256K 25408K accept 1 0:00 0.00% python 582 msf 1 760 53772K 5580K select 0 0:00 0.00% postgres 575 msf 1 760 53748K 5288K select 0 0:00 0.00% postgres 588 msf 1 760 53588K 5360K select 0 0:00 0.00% postgres 574 msf 1 760 53564K 5060K select 0 0:00 0.00% postgres 586 msf 1 760 52140K 14988K select 0 0:00 0.00% python 601 root1 40 29388K 3944K sbwait 1 0:00 0.00% sshd 604 msf 1 760 29364K 3996K select 1 0:00 0.00% sshd 578 msf 1 760 29224K 5472K select 0 0:00 0.00% postgres 576 msf 1 760 29216K 5372K select 0 0:00 0.00% postgres 583 msf 1 760 20488K 5224K select 0 0:00 0.00% postgres 589 msf 1 760 20488K 4988K select 0 0:00 0.00% postgres 579 msf 1 760 20484K 5220K select 0 0:00 0.00% postgres 580 msf 1 810 19548K 5284K select 1 0:00 0.00% postgres 584 msf 1 770 19536K 5260K select 1 0:00 0.00% postgres 590 msf 1 760 19512K 5012K select 1 0:00 0.00% postgres PID USERNAME THR PRI NICE SIZERES STATE C TIME WCPU COMMAND 2256 msf 4 200 83228K 31876K kserel 1 0:01 0.00% python 2257 msf 1 40 56340K 19836K sbwait 1 0:00 0.00% postgres 582 msf 1 760 53920K 36916K select 0 0:11 0.00% postgres 575 msf 1 760 53748K 3948K select 0 0:00 0.00% postgres 588 msf 1 760 53708K 36704K select 0 0:03 0.00% postgres 574 msf 1 760 53564K 3856K select 0 0:00 0.00% postgres 586 msf 1 760 52140K 13912K select 0 0:00 0.00% python 2641 root1 40 29388K 2876K sbwait 1 0:00 0.00% sshd 2644 msf 1 760 29364K 2904K select 1 0:00 0.00% sshd 578 msf 1 760 29224K 4140K select 0 0:02 0.00% postgres 576 msf 1 760 29216K 4048K select 0 0:00 0.00% postgres 583 msf 1 760 20488K 3940K select 0 0:00 0.00% postgres 589 msf 1 760 20488K 3808K select 0 0:00 0.00% postgres 579 msf 1 760 20484K 3896K select 0 0:00 0.00% postgres 580 msf 1 810 19548K 3960K select 1 0:00 0.00% postgres 584 msf 1 760 19536K 3936K select 1 0:00 0.00% postgres 590 msf 1 760 19512K 3828K select 1 0:00 0.00% postgres Robert ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to [EMAIL PROTECTED]
Re: Memory leak?
On Tue, 14 Feb 2006 11:00:46 +1100, Robert Leftwich I can't see anything that explains the discrepancy. Below is the top -o size after a reboot, followed by the current top after 8 datasets (the extra python process is the analysis app - at a low memory usage point): Oops, just noticed that the analysis app was at a *really* low memory usage point, i.e. it wasn't running at all, having spat the dummy on some bad data! So there wasn't any extra python app. Robert ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to [EMAIL PROTECTED]
Memory leak?
After running some number crunching for the last twelve hours I noticed my box starting to use swap. Given that it has 4gb in it (of which 3gb is available, see my other email for that issue) and I know that the app never uses more than around 1gb I was surprised. Looking at the numbers from top I was even more surprised, there seems to be a significant chunk of memory unaccounted for (from memory when I checked after a couple of hours the inactive memory was around 1300M with Free in the 400M range, everything basically totalling to around the 3gb mark as expected). The app is driven by a script and is only running for around 1/2 hr per dataset after which it shuts down and restarts on a new dataset, so all memory should be freed up/made inactive after each restart, no? Mem: 274M Active, 227M Inact, 263M Wired, 95M Cache, 214M Buf, 4536K Free Swap: 4068M Total, 707M Used, 3361M Free, 17% Inuse real memory = 3221159936 (3071 MB) avail memory = 3106529280 (2962 MB) What's the best way to track down more information as to the cause of this problem? Thanks Robert ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to [EMAIL PROTECTED]
FreeBSD 6.0, amd64, A8N-SLI and 4gb ram
I've been running FreeBSD 6.0 stable on an A8N-SLI Premium amd64 box with 2gb of ram for a few weeks now and it has been running well, but some of the analysis I run needs more ram, so I dropped in another 2 gb and started having all sorts of weird problems, such as DHCP failure and running very slowly. I tracked it down to 2 bios settings, both related to PAE (one turning on s/w PAE, the other h/w) - if i turn both of those off, then everything works, but FreeBSD can only see 3gb of ram. Some documentation suggested a custom kernel with 'options PAE' enabled would be required, but adding that generates an 'invalid option PEA' error when running make. Is it possible for FreeBSD to access the full 4gb on this m/b? Robert ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to [EMAIL PROTECTED]
Re: FreeBSD 6.0, amd64, A8N-SLI and 4gb ram
On Sat, 11 Feb 2006 23:19:07 +1100, Robert Leftwich [EMAIL PROTECTED] said: I've been running FreeBSD 6.0 stable on an A8N-SLI Premium amd64 box with 2gb of ram for a few weeks now and it has been running well, ...[snip] Forgot to mention, I'm running a kernel with 'options SMP' + GENERIC minus one or two of the non-applicable options. Robert ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to [EMAIL PROTECTED]
Re: FreeBSD 6.0, amd64, A8N-SLI and 4gb ram
On Sun, 12 Feb 2006 00:10:34 +0100 (CET), Wojciech Puchar [EMAIL PROTECTED] said: off, then everything works, but FreeBSD can only see 3gb of ram. Some documentation suggested a custom kernel with 'options PAE' enabled would be required, but adding that generates an 'invalid option PEA' ^ is it typo now or in kernel config? Typo is now (unfortunately), it says 'invalid option PAE' Robert ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to [EMAIL PROTECTED]