SSD Drive

Do you think if I install a SSD drive and have it mounted to my operating file system (/tmp /swap and /) and have my larger 1TB disk drive mounted to /home that my user's databases/apache/php crunch much faster?

(Using DA on CentOS)

Has anyone else tried this? Will this work?


I can get a SLC drive for ~$500, which would be cheaper than using raid10 in a 4 x 15k RPM SAS drive configuration.

My single MLC SSD drive on my desktop blows my raid0 10k RPM array out of the water.

You don't say what size SSD drive you can get for $500; you may need a lot of space depending on what's in your / system.

And /tmp and /var (including the logs) use a lot of space and are quite volatile. I've read that SLC drives are more reliable and faster, but I don't see any specs on write cycles before they fail.

Jeff

It's a 64GB drive. It cost about 2x more than a MLC.

I guess SSD drives in a server environment are still very new it's hard to tell how long the drive would last.

In theory, it should be able to push database queries 10x faster than a normal disk.

alphadeity said:

It's a 64GB drive. It cost about 2x more than a MLC.

I guess SSD drives in a server environment are still very new it's hard to tell how long the drive would last.

In theory, it should be able to push database queries 10x faster than a normal disk.

Click to expand...

I don't like SSD in servers, they wear out - it may take a while, but they do.

@Dougy,

Do you have personal experience, links, or only anecdotal evidence?

SLC drives are supposed to be more resistant to what you call wear (actually failure after large numbers of writes), than are MLC drives, but I'm not sure of how that translates in actual use in servers, which get a lot of constant reads.

I'd still be personally afraid to use them for /tmp and logs, though, as they get more writes than reads.

My laptop uses a 40G SSD drive, but it doesn't get a lot of use, and that's my only personal experience.

Jeff

From what I've read, CentOS doesn't natively support TRIM, which is the only thing keeping me from at least testing it out in a production environment.

SLC drives are apparently designed to last significantly longer than MLC drives. I guess being such a new technology, it's hard to say what the long term statistics would be. From what I've read, they last as long as, if not longer than a standard disk drive.

I can't really find much information about SSD and linux servers though.

I have heard that SSD slow down and wear-out fast in database/swap so until they have a common working solution I wouldn't bother with how much they are charging. Maybe a RAID1 2xSAS 15K rpm solution would work good for you for the database and keep the /home partition on a different setup?

The database runs from /var/lib/mysql in most cases so using the SSD/SAS on /home wouldn't help with mysql.

devonblzx said:

I have heard that SSD slow down and wear-out fast in database/swap

Click to expand...

SSD drives don't slow down; they continue to run at the same speed until they fail. They may appear to wear out as you put it, because after a certain number of read-write cycles to a specific address, that location will no longer be readable and will be marked as bad.

The database runs from /var/lib/mysql in most cases so using the SSD/SAS on /home wouldn't help with mysql.

Click to expand...

That's entirely OS Distribution specific.

Jeff

jlasman said:

SSD drives don't slow down; they continue to run at the same speed until they fail.

Click to expand...

They DO slow down, because they try to allocate new AND updated data to empty blocks. So each "update" is an "insert" in a empty block, followed by making the original block "invalid". But as you write more and more data, the discs eventually has to erase this "invalid" blocks, to make room for more data. This garbage collection process should run without any hassle in the background, but could possibly not catch up with the free blocks that are required at any given moment, in which the write has to wait for any free blocks created by the garbage collection process..

So write-speeds tend to become slower as you update more and more data, and the disks don't have sufficient "free" blocks, an serious problem in systems that are continuous under high-loads. Reads aren't affected by this (because SSD's don't have fragmentation problems because of their low access-times).

More information can be found at: http://www.anandtech.com/show/2738/8

Yes I did forget about the invalid block issue. You're right.

Jeff

isn't TRIM supposed to rectify this problem though?

alphadeity said:

isn't TRIM supposed to rectify this problem though?

Click to expand...

You're correct, it doesn't completely "solve" the problem, but it helps a lot when it is well implemented within your OS.