jwcolby
jwcolby at colbyconsulting.com
Wed Sep 21 15:55:21 CDT 2011
> 'We plan for the failure' LOL, and who says I am not? All I said is that there is no "server environment" but rather around a billion "server environments" all different. Do you buy an 18 wheeler because you might need to move once every 20 years? Could you afford an 18 wheeler even if you wanted to do that? Personally I rent one as necessary. For my server, I have a system with a 16 port Areca raid controller, holding 6 tb of raid 6 volumes with hot spare. These hold most of my data. My SSDs are updated once a month. They are backed up to that raid 6 array immediately after the update. If the SSD does fails I haven't lost anything. But it isn't going to fail. *EVER*, at least not from write wear. "We plan for failure" doesn't mean I have to spend money I don't have for tools I don't need. Read carefully and you will find that this specific author's opinions are not widely held, in fact I would call him borderline nutcase. Not to mention the fact that most of what he wrote was written centuries ago (in the SSD world) and just about do not apply today. I can tell he believes that stuff but I did extensive reading of white papers analyzing the problem and the consensus is that unless you are constantly streaming data to the drive as fast as the SATA port will bear, you are never going to wear out even the consumer grade drives. Instead of reading one ... person's opinion I do my own analysis. Did you know that as sectors start to have read errors the controllers move the sectors and mark them bad? This is true in rotating and SS media. So let's do an analysis (for my client building the new system). 1) With a real data set of perhaps 5 gigs 2) And a 100gb (formatted, available) ssd, if you get 5K writes for any block before you get errors This means you can move the entire 5 gigs 20 times before you can no longer write to the disk, which turns into 100 thousand writes of that entire 5 gigs of data. But if the customer is writing 5 gigs once and then updating perhaps 1 to 2 megs of data a day... This is essentially a write once database. Not quite but you get the picture Let's be brutally honest, this thing is going to last about a jillion years. Well, maybe only a million years. They are going Raid 1 with hot spare, and an intelligent controller that costs more than the SSDs. They are also buying a brand (which I recommended) which uses the 34 micron chips precisely because the new 28 micron chips are reporting all kinds of errors at this point in time. And they already (with their existing server) do a full nightly backup. So I need to tell them that the SSD has to be a thousand dollars a pop SLC drives because...???? > I think the only rational approach is expressed by 'M' at the bottom of the 'Debunking SSD lifespan and random write performance concerns' post: > > 'We plan for the failure' I think the only rational approach is 'Use my own brain'. ;) And to imply that I do not plan for failure is unkind at the very least. I have been doing business for the million a month client for 7 years and have never lost data. I have actually lost disks (two in that time period) but the hot spare dropped in and the controller rebuilt the array automatically. I have moved the array twice (to new servers) and never lost data. Raid 6 for main storage and backups has done me well. Believe me I do not have a cavalier attitude with my client's data. OTOH you seem to have a cavalier attitude about spending other people's money! ;) John W. Colby www.ColbyConsulting.com On 9/21/2011 3:26 PM, Kenneth Ismert wrote: >> jwcolby: >> ... >> 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. >> ... >> > > That seems to be a fairly cavalier attitude given the value of the data you > are handling... > > I think the only rational approach is expressed by 'M' at the bottom of the > 'Debunking SSD lifespan and random write performance concerns' post: > > 'We plan for the failure' > > -Ken