Highly Critical bug report - Rare Data Corruptions (NTFS), on SRV2003,XPx64,VistaX64 and SRV2008
dear sirs / moderators,
for few months coping critical bug. found no official way report microsoft.
therefore, asking kindly forward responsible team, core os development,
persons responsible ntfs.sys. please aware critical bug report ,
checked possible causes on end several times make sure error inside os kernel.
bug description:
- operating systems affected: xp x64, vista x64, server 2008, server 2008 (i did not check 32-bit versions !)
- hardware: more 1 cpu core seems abet actual occurrence of bug
- file system: ntfs (3.1), cluster sizes (!)
symptom:
windows corrupt data written files in rare cases when high number (> 10000) of files gets opened,
written, closed sequentially.
here more details:
i have software in place needs write ~15000 data files, in intervals of e.g. 60 minutes.
10000 (66.6%) of these files small database cache indexes, 1-5kb in size.
the remaining 5000 carry actual compressed data , 500kb - 5000kb each.
every hour, each file read, appended bit of data, re-compressed , re-written disk.
this results in 300.000 file open -> read -> write -> close operations per day.
to make short, noticed 1 out of 500.000 file write transactions in scenario write corrupted data disk. data corruption manifests in few bytes written "wrong", or few bytes "inserted" in otherwise ok looking file.
it appears issue ntfs+the kernel disk caching mechanism, because: problem _disappears_,
when call flushfilebuffers() after writes.
also, calling sleep(about 100ms) between disk writes seems prevent corruptions, uncertain.
the problem worsens much, if have, e.g. transparent disk defragmenter running. problems seems worse when running on multiple cpu cores. setting drives "optimize advanced performance" option seems worsen it.
it unrelated kind of system calls use, whether _open, _write , _close (crt) or getopenfilehandle, writefile (win32) etc.
thanks time.
for few months coping critical bug. found no official way report microsoft.
therefore, asking kindly forward responsible team, core os development,
persons responsible ntfs.sys. please aware critical bug report ,
checked possible causes on end several times make sure error inside os kernel.
bug description:
- operating systems affected: xp x64, vista x64, server 2008, server 2008 (i did not check 32-bit versions !)
- hardware: more 1 cpu core seems abet actual occurrence of bug
- file system: ntfs (3.1), cluster sizes (!)
symptom:
windows corrupt data written files in rare cases when high number (> 10000) of files gets opened,
written, closed sequentially.
here more details:
i have software in place needs write ~15000 data files, in intervals of e.g. 60 minutes.
10000 (66.6%) of these files small database cache indexes, 1-5kb in size.
the remaining 5000 carry actual compressed data , 500kb - 5000kb each.
every hour, each file read, appended bit of data, re-compressed , re-written disk.
this results in 300.000 file open -> read -> write -> close operations per day.
to make short, noticed 1 out of 500.000 file write transactions in scenario write corrupted data disk. data corruption manifests in few bytes written "wrong", or few bytes "inserted" in otherwise ok looking file.
it appears issue ntfs+the kernel disk caching mechanism, because: problem _disappears_,
when call flushfilebuffers() after writes.
also, calling sleep(about 100ms) between disk writes seems prevent corruptions, uncertain.
the problem worsens much, if have, e.g. transparent disk defragmenter running. problems seems worse when running on multiple cpu cores. setting drives "optimize advanced performance" option seems worsen it.
it unrelated kind of system calls use, whether _open, _write , _close (crt) or getopenfilehandle, writefile (win32) etc.
thanks time.
hello, marc
i forwarded issue ntfs team. time , reporting behavior us!
--malu
Windows Server > File Services and Storage
Comments
Post a Comment