Russ Allbery([EMAIL PROTECTED])@2001.05.07 00:57:48 +0000:
> Karsten W Rohrbach <[EMAIL PROTECTED]> writes:
> > Steve Kennedy([EMAIL PROTECTED])@2001.05.05 19:08:32 +0000:
>
> >> You could also use EMC storage as a back-end, not cheap but very
> >> flexible and reliable.
>
> > emc is good as long as it runs. if some pice fails those boxes are gonna
> > fsck forever...
>
> Huh? What are you talking about?
>
> About the only drawback to EMC is that it's insanely expensive. And I
> really mean it when I say insane. It's incredibly reliable, though;
> dual-redundant systems from end to end, at least in the Symmetrix. (The
> Clarion products are very different and not really the same sort of
> storage.)
emc uses a bsd based filesystem implementation. the boxes are darn
expensive since they try to make the hardware this resilient against
outages that the logical filesystem layer never catches an error. but
if it does, you're hosed. maybe you read about those meltdowns of german
webspace provider strato... they use emc.
the bottom line is, that if a disk fails nothing goes wrong, if one bus
fails nothing goes wrong, but if you got unrecoverable data errors on a
controller or other bad components, the filesystem gets damaged. due to
it's nature, being a ufs/ffs, the box comes up and check filesystems,
and you know how long that action lasts when you're checking >1tb...
i personally prefer the netapps (although the filesystems are somewhat
limited in size compared to emc or ibm) since they invested a lot of
grey matter in designing the filesystem and integrating their raid into
the filesystem. wafl has no point of inconsistency by design. the worst
thing that can happen is the log device crashing (32mb nvram card) which
means, that you inflight transactions are hosed. the filesystem on
media, in fact, is correct at any point in time. the patents for wafl
can be found on patent.womplex.ibm.com (which seems to be another site
now) and are worth reading.
/k
--
> knowledge is power. power corrupts. study hard, be evil
KR433/KR11-RIPE -- http://www.webmonster.de -- ftp://ftp.webmonster.de
[Key] [KeyID---] [Created-] [Fingerprint-------------------------------------]
GnuPG 0x2964BF46 2001-03-15 42F9 9FFF 50D4 2F38 DBEE DF22 3340 4F4E 2964 BF46
