On Fri, 16 Sep 2011, Liyun Yu wrote:

Thanks David, it is good to know this.

One of our team members suggested that the SSD also tend to be
wear out every 12-16 months. I was puzzled by that statement since
I never know if the RAM or other chip-based memory device would be
wear out electronically. That is another topic maybe.

Primer on SSD flash technology limits

all of the chip based memory devices that are able to maintain the data without power have a limited number of write cycles.

In the case of modern flash drives this is in the millions of cycles on an individual block.

older flash drives wear out faster than newer ones, MLC flash drives wear out faster than SLC drives (but MLC stores two bits per cell nstead of one and has further price advantages due to volume production so it's _far_ cheaper)

flash drives write data in chunks much larger than disk sectors (128K per chunk is common and far from the largest), if the disk doesn't buffer writes agressivly there is a symptom called 'write magnification' that makes it so that when you write one 512 byte block of data to the disk it is forced to write 128K of data to the disk. As far as I know, all modern drives have large ram buffers to limit this effect.

erasing data on a flash drive is very slow, SSDs get around this by writing a new copy of the data to a new location and erasing the old copy sometime later. All modern drives contain more flash than what you pay for so that they have a large pool of 'blank' space that they can write to (this also helps spread out the writes across more flash so that you are less likely to hit the 1m write limit on any one spot). If you are doing a LOT of writing to the SSD you can run into the situation where the drive no longer has blank spots to write the data to and it has to slow down to erase space before it can write any more. you can reduce the likelyhood of running into this by not using all the space on the disk so it has more unused space (also by picking a drive, controller card, and OS that support the TRIM command so that when you free up space on disk you can tell the drive that you don't care about it amymore and it doesn't have to preserve it)

most flash drives are slower to write to than a high performance spinning drive, but they eliminate seek overhead so they can still be a win when writing, and can be a HUGE win for random reads.


I purchased a couple dozen machines with the Intel x25E SLC flash drives about three years ago. these machines are log servers and so writing data continuously (hundreds of gigs per day) I have had a half dozen of them 'fail' on me, and from discussions on the postrgres performance list this is not an uncommon thing. In most cases the failure seems to be firmware, not storage failures, but the bottom line is they quit working. In theory when a drive wears out you should still be able to read the last data written to it, you just won't be able to write anything. In practice there are times when you just can't access the disk.

David Lang

Thanks,

Liyun


On 9/16/11 2:07 PM, [email protected] wrote:
On Fri, 16 Sep 2011, Liyun Yu wrote:

Anyone used the OWC (other world computing) SSD (6G) for RAID Array?
Any suggestions/recommendations on SSD based RAID on the market
or drives good for RAID ? Or any pointers to some reference benchmark?
Yes I do have budget issues. I also visited SSd page on Tom's hardware site
which was more than 8 months old.

I did not access the SSD market for a while. Yesterday one of our colleagues sent me a link at OWC which indicated that their drives has writing speed above 500MB for some of their 6G SSD drives. I thought SSD was at 150MB usually therefore this is very attractive to me for our app which pulls a high volume of dataflow from
the central RAID.

one thing to watch out for with SSDs, most of them do not have good behavior in the face of power failures. they all cache writes in ram on the disk and most of them have no way of getting those writes to flash if the power fails.

David Lang

_______________________________________________
Tech mailing list
[email protected]
https://lists.lopsa.org/cgi-bin/mailman/listinfo/tech
This list provided by the League of Professional System Administrators
http://lopsa.org/

Reply via email to