[rsyslog] reliability of SSD disks?

[david at lang.hm]

On Wed, 19 Aug 2009, Jeff Moyer wrote:

> david at lang.hm writes:
>
>> On Wed, 19 Aug 2009, Rainer Gerhards wrote:
>>
>>> Hi all,
>>>
>>> quick question to those in the know: are SSD disks considered reliable from
>>> an auditing (or near audit-grade) point of few? Thank to a hard disk failure,
>>> I finally got such a disk in my workstation and the performance improvement
>>> is obviously very good and creates quite a different view about the volume
>>> that rsyslog can do with "disk" queues.
>>
>> they are far more reliable than normal drives, but you would still want to
>> have a mirrored pair for true audit-grade purposes. they do wear out over
>> time (although that time is expected to be several years worth of
>> continuous write activity)
>>
>> that being said, for my normal systems I am now buying a single SSD where
>> before I purchased a mirrored pair of high-speed SCSI/SAS drives
>
> I find these claims of reliability surprising, if only due to the lack
> of soak time for such drives.  There is also no mention of the class of
> device.  Are we talking about consumer grade MLC?  SLC?  Are some
> vendors' devices better than others?  Not all SSDs are created equal.

the question of the different vendors and different models of drives 
compared to each other is something that I can't speak on. My feeling is 
that there isn't enough history to make any judgements.

however, as a class, comparing SSDs to standard rotating media drives I 
see the elimination of the mechanical portions as being extrememly 
significant. In any omputer system, the mechanical parts tend to fail 
_far_ sooner than anything else, and with no warning. high performance 
rotating hard drives also generate a _lot_ of heat, which hurts the life 
of the entire system.

the failure mode of flash is such that, in general, it will fail when you 
write to it, not when you read from it.

we are definantly in the early stages of SSDs being deployed, and I may 
find in a couple years that the drives will start to fail on me. but based 
on the knowledge available now, that seems like a reasonable risk, and the 
benifit in the meantime (much faster performance at a similar or lower 
price) makes it a reasonable tradeoff for my normal systems.

for a system with critical data on it, I would still use a raid card (with 
battery backed cache) and redundant SSDs.

as always when dealing with bleeding edge technology, you need to make 
your own risk analysis.

David Lang