Messages in this thread | | | Date | Wed, 16 Jun 1999 17:47:01 -0400 | From | Peter Rival <> | Subject | Re: 2.2.10 crashes egcs-1.1.2 runs at spurious places ? |
| |
"David S. Miller" wrote:
> Date: Wed, 16 Jun 1999 23:01:10 +0300 > From: Matti Aarnio <matti.aarnio@sonera.fi> > > After all, I have "just" 8 GB disk where egcs compile happens, > 0.5 GB RAM, and 0.25 GB swap. (Alpha 21164A/433 MHz) > > Alpha tends to obtain instability every few releases, just by chance > are you using an Adaptec scsi controller? I note also that just last > week some Compaq engineer noted on this list that he was tracking down > similar problems (in fact lockups, when using ISP1000 SCSI) during > some benchmark runs.
I'm guessing that was me. (Just my ego, I guess ;) I haven't yet had a chance to test 2.2.10 on any of my larger systems. I did test 2.2.9 plus d-axp-229-4 from Richard Henderson, as well as 2.2.10-pre3 plus some SMP compile fixes and both seemed like they should be pretty stable. I could get anywhere from a simulated 100 to 150 users on a single 2CPU system before inodes would get half-written to disk before AIM attempted to delete them. *sigh* Learning more about ext2 and device drivers than I really wanted to know ;)
Anyway, there is also a d-axp-2210 patch from Richard out there. Both of these patches are at ftp://ftp.twiddle.net/pub/rth/. But I think these are only meant as SMP patches...could be wrong tho'. BTW, like Dave mentioned, what type(s) of controller(s) are you using?
As far as my benchmarks go, well, I've kind of stagnated at around 100 users or so. After that, the benchmark can't delete a file it just either creat()-ed or link()-ed. And every time, I wind up with at least one filesystem with one half-written inode when I fsck the disks. It almost feels like the open/link is working just hard enough to snag a dentry/inode and then returning and allowing the system to continue. At a low(er) load, it seems that the system finishes the work before the benchmark tries to delete it. But when the system gets loaded, I run into this problem _every_ time. And yes, this happens just the same with and without the ext2 bugfix from sct (the problem that has "been around since Linux had a VM"). Am I on the right path with this, or am I just chasing my tail?
- Pete
P.S. "some Compaq engineer". Sheesh Dave, you make it sound like I'm official around here or something. Next thing you know, I'll be fixing code or something. ;)
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/
| |