Re: SSD --- The reliability of
Penang wrote: > Please pardon me if this topic has been discussed before, but as far > as I know about the Flash memory, they have a relatively low limit on > re-write, as compared to hard disk > > In situation where database use in concern, read and write operation > takes place very often > > I know that there are pros and cons in this issue, but basically my > question is this --- How reliable is SSD? http://www.imation.com/PageFiles/83/SSD-Reliability-Lifetime-White-Paper.pdf The last page of the report addresses "wearout" as a function of writes. As far as I can tell, they use the best kinds of writes (large sequential), to make the estimate. The flash block size is 128KB. If you want to update 4KB of that, you have to rewrite the whole thing. So relatively speaking, it is "32x" more expensive, to deal with the small sized write. To understand whether SSD may be of value to you, you need to know how much data you write per day. If it is a small amount, then it may not matter what size the writes are. If you write large amounts of data, then you'd want to understand the characteristics very well. (That means sorting all the writes into "size bins".) Before you buy anything, you should also read some recent reviews of SSD products, and some of the performance problems they can have. Don't just buy the SSD with the lowest price. You're regret it. This is a hint as to what you're looking for. There are more articles around, which test for this. http://www.anandtech.com/cpuchipsets/intel/showdoc.aspx?i=3403&p=8 If you want to build a test system for evaluation, use Intel X25-E SSDs. ******* None of the above addresses "reliability" as such. The above addresses "wearout", where you know the thing will eventually fail, as a function of how many writes you've done. Wear leveling makes the most of the write cycle limits of SLC or MLC flash chips. Reliability takes into account other things, like how often a flash chip just rolls over and dies for no particularly good reason. Take the stick of RAM in my previous computer as an example - one entire chip stopped responding, after about two years of usage. Why did that happen ? That is an example of something not accounted for as wearout - that is a device failure unrelated to usage. RAM should not wear out. And yet I've had failures at the two year mark (the stick was not abused, and the computer case is cooled well). If the chips in a flash drive do that too, then not only would I have wearout to consider, I'd have device reliability (chips dying) to add to the effect. And the only way to study that in any effective way, is to get field return data for SSD drives. And nobody gives out field return data. Paul
On Aug 12, 2:41 pm, Paul <nos...@needed.com> wrote: > Penang wrote: > > Please pardon me if this topic has been discussed before, but as far > > as I know about the Flash memory, they have a relatively low limit on > > re-write, as compared to hard disk > > > In situation where database use in concern, read and write operation > > takes place very often > > > I know that there are pros and cons in this issue, but basically my > > question is this --- How reliable is SSD? > > http://www.imation.com/PageFiles/83/SSD-Reliability-Lifetime-White-Pa... > > The last page of the report addresses "wearout" as a function of writes. > As far as I can tell, they use the best kinds of writes (large sequential), > to make the estimate. > > The flash block size is 128KB. If you want to update 4KB of that, > you have to rewrite the whole thing. So relatively speaking, it is > "32x" more expensive, to deal with the small sized write. > > To understand whether SSD may be of value to you, you need to know > how much data you write per day. If it is a small amount, then it may > not matter what size the writes are. If you write large amounts of > data, then you'd want to understand the characteristics very well. > (That means sorting all the writes into "size bins".) > > Before you buy anything, you should also read some recent reviews > of SSD products, and some of the performance problems they can have. > Don't just buy the SSD with the lowest price. You're regret it. > This is a hint as to what you're looking for. There are more articles > around, which test for this. > > http://www.anandtech.com/cpuchipsets/intel/showdoc.aspx?i=3403&p=8 > > If you want to build a test system for evaluation, use Intel > X25-E SSDs. One funny thing is --- Intel has just announced the suspension of their X25-E SSD product Reason? Technical But rumors of pre-mature deaths of the SSD may cause Intel to think twice about selling products that don't work > > ******* > > None of the above addresses "reliability" as such. The above addresses > "wearout", where you know the thing will eventually fail, as a function > of how many writes you've done. Wear leveling makes the most of the > write cycle limits of SLC or MLC flash chips. > > Reliability takes into account other things, like how often a > flash chip just rolls over and dies for no particularly good reason. > Take the stick of RAM in my previous computer as an example - one entire > chip stopped responding, after about two years of usage. Why did that happen ? > That is an example of something not accounted for as wearout - > that is a device failure unrelated to usage. RAM should not > wear out. And yet I've had failures at the two year mark (the > stick was not abused, and the computer case is cooled well). > If the chips in a flash drive do that too, then not only would > I have wearout to consider, I'd have device reliability > (chips dying) to add to the effect. And the only way to study that > in any effective way, is to get field return data for SSD drives. > And nobody gives out field return data. > > Paul