You know flash is king when disk giant Seagate grows its SSD line Seagate is going to expand its solid state drive (SSD) line this year using co-developed Samsung controller technology and introducing its first multi-level cell drive. Seagate and Samsung have a flash chip supply and controller partnership. Stifel Nicolaus analyst Aaron Rakers has talked to Seagate execs and gleaned that: …
just out of personal interest : Is anyone using SSDs on enterprise applications... If so can I presume that they are only within RAID 5 bays ?
After reading a couple of articles on the manner in which SSDs die, no warning, messages, nothing, nada, I am much to scared to deploy them on anything other than non vital servers, which don't amount to being very much in my workplace.
No warning - really - try using SMART - it reports the amount used / remaining so if anything they are far more reliable and their life is typically far more predictable.
We typically use them mirrored but that's more to do with the fact we are paranoid about downtime - we even mix part worn SSDs with new(er) ones so they should not fail at the same time. Would recommend Intel SSD drives and ideally ones with super capacitors - especially in database applications.
<quote>No warning - really - try using SMART</quote>
I have been led to believe that when SSDs fail, they simply do just that they Fail. In this case SMART would not be on any advantage as the SSD would be dead.
I do not use them, hence the reason I asked.. it was an honest question
Haha! We have a server with 2x Intel 520 240GB SSDs in RAID0.
Then again, it's only a virtual desktop machine for unimportant things and has no actual critical data on. So, there's that. But for what it's worth, that's been running, loaded with virtual machines, just fine, a year in February.
We run 11 SSDs in a RAID 5 array here. They're HP enterprise models. Yes, they do generally "just die" but then in my experience, normal hard drives do that too in servers. ie. you get a message on your monitoring system "Drive 7 has failed" and that's it. In a loud server room, you're not gonna hear any noises, and I've yet to have SMART warn me of anything likely to happen.
But then that's one of the reasons to run them in a RAID array. So long as you get SSDs that support it, with TRIM.
I've already replaced the boot HDD in my 2 year old laptop with an SSD, new lease of life, produces less heat so no fan kicking in all the time, everything is faster, and the battery lasts and extra hour or more (was about 2.5 hours normally, now 3.5 to 4 hours on a charge).
Cost wise, HDDs hit the $1 per GB in 2005, SSDs hit this price early to mid last year. 2005 was also the year the first 500GB HDDs came out and we did get 480/512 SSD drives last year, and 1TB drives are now on their way. So seems SSDs are where HDDs were about 7 years ago, but SSDs are moving faster, so hopefully shouldn't be 7 years to get to a 1TB SSD for £45, or 2TB for £65, maybe 2-3 years?
I think the $1 level was the critical mass point for HDDs, suddenly people were buying more and larger drives which increased demand, increased production, which brought down costs, and so on. SSDs seemed to be going the same way till the hiccup in the market last year, but hopefully is back on track again now.
Although I don't have any direct experience with this, it seems that a lot of databases aren't well suited to using flash storage due to them not being optimised for that medium. The problem lies in the way that inserts and updates often have to make several updates on the on-disk indexes (B-trees or whatever) and each random write requires a read-blank-rewrite cycle on an entire disk block. Reads from the database, on the other hand, is something that does suit flash well since seeks are effectively free.
I don't have a link to any recent papers to hand, but if you search for "log structured database flash" you should turn up a few. The main advantage of log-structured databases is that inserts and updates only have to write once to the disk (with periodic rewrites for garbage collection to coalesce partially empty blocks). Thus you get very good write speed and since you're not going through as many of the read-blank-rewrite cycles that are typical of B-tree style indexes you should be able to extend the life of the disk by three times or more. Have a search for Fast Array of Wimpy Nodes (FAWN), too. It's slightly old, but it gives a good demonstration of the kind of speedups that log-structured databases + flash storage can achieve.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
