On Sun 23 Apr, 2000, "Shahryar G. Hashemi" shahryar@n2h2.com wrote:
All in all I do not believe that using filesystem files is a solution for ASE on any platform. With ASE 12.0 they now switch to D_SYNC [wait until the write has been comitted to the file before continuing on] for filesystem files. This of course will make using filesystem files more of an option *but* it will also slow things down [you won't take the OS at it's word anymore] since you will guarantee that data will be written to your filesystem devices.
Well, DBA acquaintances always say that Sybase don't support recovery from non-raw partition ASE setups and so won't run Production databases on anything but.
I think, regarding D_SYNC, that perhaps you mean O_DSYNC? Don't know how that plays out with NFS, as (for version 2) NFS operations are synchronous anyway (but the Filer's NVRAM allows it to return completion before data is resident on disk.) NetApp's Tech Note document specifically uses Version 2, though not apparently for reasons of data integrity (Can the author of that Tech Note, or anyone else, comment on this?)
I must add that I'm interested to see that Sybase now support ASE on Linux for production use, as of release 11.9.1, if you part with the appropriate sum of money.
On Mon 24 Apr, 2000, "Bruce Sterling Woodcock" sirbruce@ix.netcom.com wrote:
Since you're fond of testing, why don't you test this yourself?
(a bit harsh, but I agree with Bruce that if you've got your test-rig then it seems like you're probably in the best place to do the testing with your Linux system.)
James.
I think, regarding D_SYNC, that perhaps you mean O_DSYNC? Don't know how that plays out with NFS, as (for version 2) NFS operations are synchronous anyway (but the Filer's NVRAM allows it to return completion before data is resident on disk.) NetApp's Tech Note document specifically uses Version 2, though not apparently for reasons of data integrity (Can the author of that Tech Note, or anyone else, comment on this?)
This is a good point, but I didn't want to comment on it before because Mr. Hashemi says he eventually got Linux configured in a way that corresponds to the tech note, and I didn't want to question that. He he said he rebooted the client and that this corrupted his db over NFS ("filesystem") but apparrently not local disk, I concluded that it must be an issue particular to how Sybase handles now-raw partitions that was inconsistent. But if he was using NFS3 without synchronous writes, then that too could have caused the corruption. Keep in mind that this is entirely a client side issue, and nothing to do with Netapp (in that case).
The solution is clear, of course, that either Sybase supports synchronous writes for use with NFSv3, or NFSv2 should be used. I know personally that Sybase worked fine over NFS v2 without this sort of cuorruption. Then again, there could be something funky with Linux, which has had a notoriously bad NFS history.
Bruce
-
Bruce Sterling Woodcock -
jrg＠acm.org