Re: [Jfs-discussion] performance probs - 2.4.28, jsf117, raid5
On Mon, 6 Dec 2004 17:37:30 -0500, Sonny Rao wrote: Right, so there's really only one thing i can think of and it's not much of a solution. You can change the memory split so that you can use all 2gb for kernel memory. I know there are some patches floating around to convert the 1GB to a 2GB split, or you can use one of the so-called 4GB/4GB kernels which keeps the kernel in a totally different address space. This is separate to the option for high-mem support in the kernel? OK, I've just found the posting from Ingo Molnar. It's really only a solution if all of the inodes in your working set fit into 2GB, otherwise you're just delaying the inevitable. Ultimately, this is what 64bit machines (with a lot of ram) are good for :-) Yeah ... thanks again. I've managed to tweak our setup, which has helped - but the problem is that kswapd when under heavy load manages to bring the machine to a halt as other processes appear to be spinning. I'm now contemplating offloading a lot of this IO to another box - fortunately 32bit boxes are cheap to come by these days. /Per Jessen, Zurich -- regards, Per Jessen, Zurich http://www.spamchek.com - let your spam stop here! ___ Jfs-discussion mailing list [EMAIL PROTECTED] http://www-124.ibm.com/developerworks/oss/mailman/listinfo/jfs-discussion
Re: [Jfs-discussion] performance probs - 2.4.28, jsf117, raid5
On Sun, Dec 05, 2004 at 08:41:35PM +0100, Per Jessen wrote: On Sun, 05 Dec 2004 18:40:58 +0100, Per Jessen wrote: I do a find in a directory that contains 5-600,000 files - which just about makes the box grind to a halt. The machine is not heavily loaded as such, but does write 2 new files/sec to the same filesystem. Or tries to. I need to add - at the same time kswapd is very, very busy, despite only about 1Gb of the 2Gb main core being used/active. /Per Yes, this is a consequence of the way memory is partitioned on IA32 machines (which I'm assuming you're using). If you look at the amount of memory being used by the kernel slab cache, I'd bet it's using much of that 1GB for kernel data structures (inodes, dentrys, etc) and whenever the kernel needs to allocate some more memory it has to evict some of those structures which is a very expensive process. Look at /proc/slabinfo and add up the total number of slabs. Sonny ___ Jfs-discussion mailing list [EMAIL PROTECTED] http://www-124.ibm.com/developerworks/oss/mailman/listinfo/jfs-discussion
Re: [Jfs-discussion] performance probs - 2.4.28, jsf117, raid5
On Sun, 05 Dec 2004 18:40:58 +0100, Per Jessen wrote: I do a find in a directory that contains 5-600,000 files - which just about makes the box grind to a halt. The machine is not heavily loaded as such, but does write 2 new files/sec to the same filesystem. Or tries to. I need to add - at the same time kswapd is very, very busy, despite only about 1Gb of the 2Gb main core being used/active. /Per -- regards, Per Jessen, Zurich http://www.spamchek.com - let your spam stop here! ___ Jfs-discussion mailing list [EMAIL PROTECTED] http://www-124.ibm.com/developerworks/oss/mailman/listinfo/jfs-discussion
Re: [Jfs-discussion] performance
One reason may be the location of the partitions. A low sector located partition is faster than one that lies at the end of the disk. Disks usually start counting outside Sean Neakums schrieb: Hi, I've been playing with JFS this past day or so, and I am observing a performance problem. I am using a patched kernel (version 14a of Rik van Riel's rmap VM), so this may be implicated somehow. When I do a build of the LNX-BBC (http://www.lnx-bbc.org/) I get a build time of about 75 minutes on an ext3 volume, and about 84 minutes on a JFS volume. I'm using stock ext3 and JFS version 1.10, on Linux 2.4.19 plus the rmap patch. Here's some data, from time make install: On ext3: real76m37.636s user38m59.370s sys 18m34.210s On JFS: real84m13.123s user39m3.020s sys 18m49.810s The machine in question is an SMP box with 1.13GHz P-III, 256M of RAM and IDE disks. I use ccache (http//ccache.samba.org/) to do these builds, and see almost identical hit/miss statistics for each run. I believe that, due to ccache, the build becomes fairly I/O-bound. Judging by fact that the wall-clock time shows the only big variation, I'm guessing (wildly, with no proof) that this may be something to do with how JFS schedules I/O. If there is any other information you'd like me to gather, please holler. ___ Jfs-discussion mailing list [EMAIL PROTECTED] http://www-124.ibm.com/developerworks/oss/mailman/listinfo/jfs-discussion
Re: [Jfs-discussion] performance
commence Robert K. quotation: One reason may be the location of the partitions. A low sector located partition is faster than one that lies at the end of the disk. Disks usually start counting outside This occurred to me too, and so today I created an ext3 filesystem on the same volume, and redid the build. I got the same time as I did for previous ext3 builds, using the same chunk of the disk as with the on-JFS build. -- / | [|] Sean Neakums| Questions are a burden to others; [|] [EMAIL PROTECTED] | answers a prison for oneself. \ | ___ Jfs-discussion mailing list [EMAIL PROTECTED] http://www-124.ibm.com/developerworks/oss/mailman/listinfo/jfs-discussion