jwcolby
jwcolby at colbyconsulting.com
Tue Sep 20 14:54:29 CDT 2011
LOL, everyone has an angle. One of the SSD myths is that your SSD will die in a month. It actually is not a myth, it will die in a month (or a few months) if all you do is run random write disk tests for that entire month. > Don't expect consumer-level SSDs to last very long in a server environment. What exactly is a "server environment"? I have a server with 400 gigs of (consumer level) SSDs with SQL Server data files on them. I read from them all day every day. I write to each file once a month for a monthly update. That is a server environment. You should see my server chug. I do millions of dollars a month business for my client. Not my share unfortunately. The specific client I have been discussing that is building the server is a small company with about 5 gigs of data (currently). It took them 8 years to get to 5 gigs of data. If they add another 5 gigs every year for the next 10 years... (and they won't) They are a call center so they have 25 people in the database all day every day. They ask the caller for enough info to search for a claim. They open a tabbed form and can see every part of the claim. They discuss the claim with the caller. They write a couple of paragraphs about the call into a memo field which gets written to disk. They change a phone number here, an address there. That is a server environment. 25 users in the database all day! You should see the reports they generate at the end of every month for their clients!!! All read-only btw. My point Ken is that there is no such thing as "a server environment". Like anything else "it all depends" is pretty much the answer. There are as many "server environments" as there are servers. Some would kill an SSD in a month, most will never kill an SSD in the 5-10 years the server will exist. I work for small companies exclusively. All of my posts have discussed these details in excruciating detail so this is not new to anyone here. And yet what you are discussing is some "theoretical server environment" where the server is writing every block 100 times a day. I am sure that environment does exist, but it is way less common than environments like mine. And then there is this. One of the largest internet companies on the planet is buying SSD literally by the ton. http://www.datacenterknowledge.com/archives/2011/03/10/facebooks-appetite-for-ssd-boosts-fusion-io/ I'm guessing they have the expertise to analyze the cost / benefit / probability of failure. > If you do spring for them, get the (much more expensive) server-rated SSDs using SLC memory I am absolutely comfortable with the consumer grade SSDs in my server here at my office. I doubt seriously it will *ever* wear out. Likewise for the server being built at my client. In fact in these situations the life expectancy of the SSD is probably realistically 100 times longer than an equivalent rotating drive. No moving parts, very little heat, never turned on / off, a few hundred megs of writes a day. What's to wear out? > Don't expect consumer-level SSDs to last very long in a server environment. Why in the world would I pay 5X the dollars for 10 times the life span when I expect to be dead (and I'm not *that* old) long before my drives wear out? Engineering is all about *requirements* analysis. John W. Colby www.ColbyConsulting.com On 9/20/2011 2:45 PM, Kenneth Ismert wrote: > All, > > SSDs sound ideal for server applications, except for one nagging question: > when will it wear out? > > SSD endurance summary > http://www.storagesearch.com/ssdmyths-endurance.html > > > But, there are circumstances where SSDs make a compelling argument for > adoption: > > SSDs replacing HDDs? that's not exactly the way it happened > http://www.storagesearch.com/bitmicro-art3.html > > > If you do spring for them, get the (much more expensive) server-rated SSDs > using SLC memory -- see response to Micker in comments: > > Debunking SSD lifespan and random write performance concerns > http://maxschireson.com/2011/04/21/debunking-ssd-lifespan-and-random-write-performance-concerns/ > > Don't expect consumer-level SSDs to last very long in a server environment. > > -Ken