From Bugzilla Helper: User-Agent: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:0.9.2.1) Gecko/20010901 Description of problem: Client has an application that builds (then deletes) a directory tree (5 layers deep) with some 85000 directories and about 100 times that number of individual files. Attached perl script triggers a crash -- eventually, but running "bonnie++ -s 12g:1k -n 850000:1g:1k:85000" seems to reliably crash the system. Seems to crash shortly after system needs to page (swap is 2GB). Test is run on ext3. Machine is Dell 2650 w/PERC/3Di; 5 63GB drives in RAID5 config; 6GB RAM; 2 Xeon CPUs @ 2.2GHz; HT disabled in BIOS. Crash display not available (damned ACPI). Version-Release number of selected component (if applicable): kernel-2.4.20-9bigmem How reproducible: Always Steps to Reproduce: 1.Install 9. 2.Apply all updates. 3.build bonnie++-1.03a. 4.run: "bonnie++ -s 12g:1k -n 850000:1g:1k:85000" on default ext3 fs (/home). Actual Results: Crash -- almost as if "halt" command executed. Expected Results: test completion Additional info:
Created attachment 91567 [details] perl script to cause crash
Ok, after some run-around from Dell, the bottom line is that with intense filesystem activity (not just blasting i/o), even with the card's cache for the data partition in write-thru mode (as recommended by Dell), with enough i/o load, the card will lock up, never to answer to the kernel again -- at least until a hard reset.