> I had about 250GB of real files (over 400GB in the virtual pool). I never had 
> any disk corruption.

In my case it didn't show up until i had about 4T in the virtual pool, with
about 350GB of real files. 

I also noticed somewhere along the line that i had LOTS of identical
small files from someone's logging project gone awry; thousands of them on
one server.  That had to up my link count.

> => I'd blame my hardware if i hadn't had similar experiences several years
> => ago (on solaris) with an application that was written to use flat files
> => as a "database".
> 
> I've administered Solaris for over 12 years, and have never seen the kind of 

I've administered solaris for about 16 years.

> filesystem corruption that you describe--aside from hardware probelms. I'd
> seriously examine your hardware (not just the disk subsystem, but memory as 
> well).

I've seen it twice now myself.  In the first instance I am 100% sure it
was not the hardware.  This time it's possible, i haven't exhaustively
tested it; but it's not crap hardware.

> => the filesystem just didn't like lots and lots of small files and the
> 
> Huh? UFS is fine with "lots and lots" of small files--and performs much better
> than ext[23]fs under similar circumstances. The filesystem workload you're 
> describing is typical of some mail servers, and many NNTP servers--both common
> applications for Solaris.

You're certainly right; lots of people use solaris on these kind of loads.
All I know is that twice now in the described circumstances I've had
problems.  Actually, I can think of a third instance, when a problem
like this was reported on the openafs list by someone using the namei
fileserver on solaris.

and under extreme load it apparently dies.  I can't tell you why, I
can just tell you what we observed.

I should add that in the previous instance where i ran into it, I was the
junior admin and the senior guy had about 20 years' experience.
Unfortunately we couldn't replicate it reliably, and I can't do so
this time.

> => system panicked every few weeks, and fsck took FOREVER.  I'm guessing
> 
> Well, yes, fsck will take a very long time. Do you have journalling turned on 
> as a filesystem mount option?

Of course I have journalling turned on (if i recall correctly, it's
the default on solaris 10).  the filesystem needs an fsck to repair
the corruption.

Typically it's been "unexpected unallocated inode" errors. 

> What is the average size of your files? Have you tuned the file system (block 
> size, number of inodes, etc.) for that type of file?

I'm using the default block size; I have plenty of inodes.  

> Um, as far as I know, ZFS has been in general release for over a year 
> already, (and in 
> limited release since late 2004).

Yup.  And in my book, that's not mature.

Go read the zfs list on opensolaris.org.  Some folks are extremely happy;
others have trouble.  Bugs are getting filed (and fixed, I'm sure; I'm
just afraid of the bleeding edge.

> I want a newsgroup with a infinite S/N ratio! Now taking CFV on:
> rec.motorcycles.stagehands.pet-bird-owners.pinballers.unix-supporters
> 15+ So Far--Want to join? Check out: http://www.panix.com/~bergman 

Sorry, I don't own a bird.  And I haven't been a stagehand for 20 years
although i'm sure that won't stop me from pithy ruminations.

danno
--
Dan Pritts, System Administrator
Internet2
office: +1-734-352-4953 | mobile: +1-734-834-7224

-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys-and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
_______________________________________________
BackupPC-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/backuppc-users
http://backuppc.sourceforge.net/

Reply via email to