I am going to build a new server to handle a high-traffic php/mysql site. I am going to use SSD disks to benefit from fast random read performance but really got confused about choosing the right SSD type and (possible) RAID arrangement.

I need a degree of redundancy so tend to use disks in RAID 1, however, I am not sure whether TRIM will be possible in RAID 1 in my chosen RAID controller, and if TRIM is not possible in R1, what would be the impact on life expectancy of the disks. (I've read that not using TRIM causes the SSD performance degrades due to excessive writes)

Here are my (tentative) hardware specs:

Intel Xeon E3-1270 CPU
Supermicro X9SCL Motherboard
3Ware 9750-4I SAS/SATA RAID Controller Card
2 (?) x ~120GB SSD (NOT DECIDED YET!)

So I appreciate your hints to choose an optimal disk model and a suitable RAID arrangement.

  • You could use software-RAID to ensure the SSDs see TRIM requests. This has the side-bonus of saving you the price of the card, which you could apply to "better" (or, more spare) SSDs. – mlp Aug 11 '11 at 2:36

As far as I know, TRIM won't be passed to SSDs in hardware RAID. It's not so much that the lack of TRIM will affect the lifetime of an SSD (although this may be a factor), but that the drive's performance will suffer relatively quickly - which somewhat contradicts the purpose of the SSD.

If you really need RAID 1, then I'd favour SandForce-based SSDs, which tend to cope better without TRIM.

Edit: It's difficult to quantify how much the MTBF would reduce without TRIM, without being an SSD manfacturer. But we do know that the lack of TRIM can increase write amplification, thereby shortening the life of the drive due to extra program-erase cycles. SandForce drives have really low write amplification rates anyway - typically less than 1.0, due to compression - which explains why they cope better than other SSDs without TRIM.

It's also important to clarify that there are two types of SSD failure:

  1. Wear of the flash memory due to excessive program-erase cycles over time
  2. Failure of the SSD controller hardware or software

It is most likely the second of these which is described here, and which accounts for the relatively high failure rates here (albeit, presumably for consumer drives). In these cases, because failure occurs relatively early, it's unlikely that TRIM would make any difference to the reliability of the drive.

  • Do you know any benchmark to indicate negative effect of not using TRIM on SSD MTBF? – alfish Aug 10 '11 at 6:43

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.