Thanks for sharing  your experience.

[email protected] writes:
> in terms of a flash-based cache, I would have some concerns.
> 
> one is the durability of the flash, with only 1G of flash heavy writes
> are going to wear it out much sooner than the same amounts of writes
> to a much larger SSD with wear leveling.

I would be completely fine with a 3 year lifespan.  (granted, the
thing is going to get /thrashed/ for those three years, so it still
may be a problem;  but this isn't something that needs to last for 
5 or 10 years.)  I'd also be buying the X25-E SSDs.  no point
in messing around with the cheap stuff when it's cache, which 
to some extent ameliorates your second concern.  I'd also mirror
two SSDs, most likely.  Of course, if they both wear out at the same
time, that doesn't help, but if there is an element of randomness,
it may help quite a lot.  

> another is the write speed, flash is relatively slow to write to, I
> would not expect it to have nearly the performance of battery-backed
> ram.

This is true, but it's ridiculously better than disk, at least
the good SSDs are, especially for random writes, and when you cram
a bunch of virtuals on a server with four spindles, it's all random. 

>From my point of view, more slower cache is much better than less
faster cache, just because my spinning disk is positively glacial
when it comes to random access.

> hardware raid (with battery backup) is significantly more durable,
> with software raid you have a problem that the OS may have updated
> some of the drives, but not others at the time you loose power, so the
> stripe may be in an inconsistent state.

Hm.  I use raid 1+0 or raid10 in software raid, only.  the 
performance characteristics inherent to RAID5 without a battery backed
cache make it untenable for my use, before looking at reliability.

I've long experience with raid1 on md, and I've never had a problem
with losing power (sometimes it requires a raid rebuild, but no
lost data.)  I've recently switched to raid10 (rather than striping
across to raid1 devices)  and I wonder if that will change this
record.

> the battery backed cache can also give you huge performance
> improvements for workloads that do fsyncs to make sure the data is
> safe on disk (mail servers and database servers commonly do this) as
> they usually are faster than doing the same I/O without a cache and
> without fsync.

Yeah. persistent cache, for me, is the primary reason I'm thinking
of hardware raid.  My understanding is that it can take a big bite out 
of the read-calculate-write overhead of raid5, which gives me more 
spindles for reads (caching doesn't help nearly as much for reads, when
you have a large amount of data being read, relative to cache size) 


My understanding is that persistent disk caches also help make random
writes a little bit more sequential.  
  
> the failure modes you run into with hardware raid are also frequently
> much nicer than with software raid. If the raid card has problems
> reading a drive, you still get your answer from the other drives at
> the normal degraded performance. But if the kernel has problems
> reading a drive, it keeps retrying, which can take quite a while,
> during which time lots of things just stop.

This has not been my experience, at least not when I tested a 
'half dead' consumer grade drive that sat and retried all day in md
in an expensive 3ware raid card.   The damn think kept retrying
(it was a mirror member) bother in the 3ware and in md.  Switching
to 'enterprise' sata solves the problem at a cost of paying 50%
more per disk, but that's true of both MD and 3ware.  

-- 
Luke S. Crawford
http://prgmr.com/xen/         -   Hosting for the technically adept
http://nostarch.com/xen.htm   -   We don't assume you are stupid.  
_______________________________________________
Tech mailing list
[email protected]
https://lists.lopsa.org/cgi-bin/mailman/listinfo/tech
This list provided by the League of Professional System Administrators
 http://lopsa.org/

Reply via email to