TRIM is released...

Silencing hard drives, optical drives and other storage devices

Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee

cmthomson
Posts: 1266
Joined: Sun Oct 09, 2005 8:35 am
Location: Pleasanton, CA

Post by cmthomson » Sat Oct 31, 2009 3:22 pm

Even though I have a G1 SSD, meaning TRIM wouldn't help me, I downloaded the tool to see what SMART would show.

Sure enough, it had some useful information: it provides the total data write volume, and also the "media wearout indicator". In my case, my year-old drive has accumulated 5.69 TB over 7819 hours (17.5 GB/day), and has a wearout of 93. If this is to be believed, my drive is good for about 10 more years...

TBorgeaud
Posts: 13
Joined: Fri Sep 05, 2008 2:31 am
Location: London, UK

TRIM and hype

Post by TBorgeaud » Tue Nov 03, 2009 4:01 pm

I can't help feeling that the arrival of TRIM has been accompanied with a certain about of hype.

Is it just me that thinks that TRIM is rather less of a solution to a problem and more of a nice performance optimization or work around for a particular aspect of SSD behaviour.

It seems to me that while there may be advantages gained in making a storage device aware of file system operations, to at least some degree, there are also disadvantages. Having to inform a drive about logical blocks no longer in use surely means that the drive can no longer be treated as a simple, independent persistant store of blocks.

This might not be a big disadvantage, but I imagine that certain assumptions valid for most conventional block level storage devices would no longer be valid and that enabling TRIM to work through various layers of virtual disk devices (i.e. various sorts of volume management) may not be easily possible.

I think I'd rather live with the performance that results when the contents of unused filesystem blocks is shuffled around. Am I right in thinking that the typical performance of current SSDs that no longer have huge pools of unused (according to the SSD) blocks is still very much better than most conventional hard disks?

Or, is it the case that the performance of an SSD with plenty of free blocks makes if feel completely different to a device without such a pool of free blocks? How noticable is the drop in (system) performance that occurs without TRIM?

nutball
*Lifetime Patron*
Posts: 1304
Joined: Thu Apr 10, 2003 7:16 am
Location: en.gb.uk

Post by nutball » Wed Nov 04, 2009 2:07 am

I don't agree, I think TRIM is pretty important.

You're right that it currently doesn't play very nicely with logical volume managers and RAID arrays, but to my mind that just means that those bits of the software/firmware stack need to catch up. Clearly this will take some time, but SSDs are here to stay for the forseeable. Even if TRIM didn't exist there's quite a lot of software which will likely need re-engineering to get the best out of SSDs - I'm think file systems, DBMSs, etc., which have been heavily optimised over the years to deal with the performance traits of HDDs and now seen to be sub-optimal for SSDs.

TBorgeaud
Posts: 13
Joined: Fri Sep 05, 2008 2:31 am
Location: London, UK

Alternative to TRIM

Post by TBorgeaud » Wed Nov 04, 2009 4:49 am

A second question (regarding the importance of TRIM):

If the apparent (usuable) capacity of an SSD is somewhat smaller than the actual available (physical) blocks, such that the additional pool of reserved blocks or sectors is large enough, would that not result in avoid the kind of performance degredation that the use of TRIM addresses?

I imagine that there would still be redundant shuffling of blocks which may actually be unused by filesystems (or similar), but would there not be much less need to shuffle blocks to accomodate newly written data (i.e. on less worn areas of the SSD)?

Of course, there would be the cost that giving a device a larger reserve would mean that the device provides a smaller usuable capacity.

However, this disadvantage might be offset by not having to worry about making sure that TRIM is used correctly.

Or, would there still be significant disadvantage in not informing a device about blocks which it could treat as unused?

Erssa
Posts: 1421
Joined: Sat Mar 12, 2005 9:26 pm
Location: Finland

Re: Alternative to TRIM

Post by Erssa » Wed Nov 04, 2009 10:14 am

TBorgeaud wrote:A second question (regarding the importance of TRIM):

If the apparent (usuable) capacity of an SSD is somewhat smaller than the actual available (physical) blocks, such that the additional pool of reserved blocks or sectors is large enough, would that not result in avoid the kind of performance degredation that the use of TRIM addresses?

I imagine that there would still be redundant shuffling of blocks which may actually be unused by filesystems (or similar), but would there not be much less need to shuffle blocks to accomodate newly written data (i.e. on less worn areas of the SSD)?

Of course, there would be the cost that giving a device a larger reserve would mean that the device provides a smaller usuable capacity.

However, this disadvantage might be offset by not having to worry about making sure that TRIM is used correctly.

Or, would there still be significant disadvantage in not informing a device about blocks which it could treat as unused?
I think Anandtechs SSD anthology can aswer your questions.

http://www.anandtech.com/storage/showdo ... i=3531&p=1

Post Reply