On 03/26/2007 08:01 AM, Andy Speagle wrote: > > I have no idea why you are getting any lockups as a result of software > raid as I have over 7TB of data stored on linux software raid (5 and > 6) and I have never experienced a single lockup from that although > most of this is on server class systems I do have a few systems on > older desktop athlon systems using either highpoint or promise > controllers and 120GB WD drives. I am using gentoo on and 2.6 kernels > on all systems and in most cases I am using reieserfs with a few being > XFS or ext3. Also most of my raid arrays have lvm on top of md as I > try to avoid having a single filesystem of > 1TB. > > > As for your question of iSCSI I believe that would totally eliminate > the problem but it will be very costly. Are you using the onboard ide > controller or some other card? Have you ran memtest86 for at least 24 > hours with your system? Do you have smart running on your drives? If > so have all the drives passed? Are there any io errors in your dmesg > output? > > > > This is most definitely a desktop class system that I'm using. I have > not run an extended memtest as you suggest, I do have SMART running on > the drives (4 x 250GB ATA-100) and they seem to be healthy. I'm using > the on-board controller with an add-on PDC20248. I too am using LVM > on top of my RAID arrays. > > I'm wondering however, if I'm using my storage inefficiently... I have > the disks partitioned as such: > > /dev/hdx1 - 128MB > /dev/hdx2 - 100GB > /dev/hdx3 - Remaining Space > > I have configured /dev/hdx1 (md0) as my RAID1 /boot partition outside > of LVM... my /dev/hdx2 (md1) as a RAID-5 array as a PV for my root > VG.. system partitions... and /dev/hdx3 (md2) as a PV for my Video > VG... where I store everything. > > The lockups I have... which are more infrequent now since I've dropped > the buggy VIA motherboard... are ALWAYS related to the Software > RAID... because when I bring it back up... md2 is dirty and must > resync... > > I have a pretty hefty power-supply in the system... in my > estimation... it's no smaller than 450W... which should be enough, I > thought. I'm just annoyed with the whole thing and am looking for a > solution. > > Thanks, > > Andy
Hi Andy, There are kernel stack overflow issues with LVM -> RAID5 -> XFS if your kernel is using 4k stacks. RAID1 is okay. Newer kernels (2.6.20) are better, but the real problem is with XFS using too much stack space. They're working on it (see xfs mailing list). The most solid workaround is to recompile the kernel with the old 8K stack (which also pushes interrupt queues out of the stack). Unfortunately, this means you can't use prebuilt kernel binaries any more. I had a few systems doing just that (which were hard-hanging even with Reiserfs) under heavy I/O and been rock solid since. Shouldn't affect you badly as you're probably not runnning gazillions of threads (Java threads often come up in this discussion). Let us know how you end up. Regards, Chris PS - I would have partitioned just, say, 20GB RAID1 (gives good read performance) or RAID5 (slower writes) for hdx2 / root filesystem, leaving the rest RAID5 (for capacity) LVM on hdx3. You didn't say what RAID you used for hdx3/md2. _______________________________________________ ivtv-users mailing list [email protected] http://ivtvdriver.org/mailman/listinfo/ivtv-users
