2

Does anyone have any experience with 3rd party SSD drives in any of the Dell Blade range?

I'm not prepared to pay dell £1000 for a 100GB SSD drive, when I can pick up 2x 256GB drives for £800!

I've read this : Dell SSD options and 3rd party SSDs in Dell hardware but it doesn't seem to mention anything about the raid controller, or if the new drives will simply slot into the SAS/SATA enclosure.

Community
  • 1
trm201
  • 21
  • 1

2 Answers2

1

When you purchase SSDs you should be aware that there is a major technology difference between consumer grade and "enterprise" grade SSDs. I wouldn't recommend MLC for buisness production environments.

Consumer grade SSDs will use MLC (Multi-Level Cell):

  • Lower Write Performance
  • Faster wear over time

Enterprise SSDs are SLC (Single-Level Cell):

  • Better Write Performance
  • Lower wear rates

I believe the crucials are all MLC so you probably want to go a SLC SSD drive like the Intel-X25E. I can't tell you how they work in blades, but if you watch the Server Fault blog you should see how they perform in a Dell R710 in the next couple of weeks. We have an H700 controller and I believe the firmware will no longer complain about 3rd party drives -- but we will find out soon.

As far as the SATA/SAS it will depend on your backplane. The R710 will support either SATA or SAS with the same backplane as long as they are not mixed (Might be workarounds -- but not Dell recommended). I would call dell and ask about your particular blade server when it comes to this.

Kyle Brandt
  • 82,107
  • 71
  • 302
  • 444
  • Really helful answer, thank you Kyle. Am I right in thinking that the wear rates simply mean that the consumer level drives will refuse to write any more data earlier than the enterprise level drives will? Or is there something more sinister/serious to bear in mind when considering using the cheaper disks in a server? – Stephen Kennedy Feb 20 '12 at 15:27
  • This should actually be updated... some enterprise drives (the newest Intel 710 we just put in our DB Severs are MLC). But enterprise drives seem to have less firmware bugs and I would put more trust in them. When an SSD fails it generally won't do so gracefully (in particular, it won't be predicted behind many raid cards these days). RAID will help, but personally I only really trust the best SSDs even in raid for critical servers. – Kyle Brandt Feb 20 '12 at 16:07
1

I think the answer for the performance issue is twofold: First, I own one of the C300 (265 Gb) SSDs, 

I use hdparm to secure-erase the drive, and I hot-plug because the unlock scripts don't seem to work with the controller? ioctl() I/O error.

(Hint don't use Windows, it is BRUTAL to the drives. It only exists to make the COMP-TIA racket more profitable.) fdisk the drive with -b 4096 with the CHS data < hdparm -i [node].

I use Mark Lord's wiper.sh that comes with the source code version of hdparm[v.9.37] that TRIMs the free space, BUT requires a partition file system, or it will abort.

THEN (3rd step) I use gparted on KNOPPIX 6.4.4 to offset the first partition by 4MB, and thereafter align by the MB, with 4MB separations, where partitions are multiples of same 4MB. (not hard at all, obviously.) I bear you witness that my MLC EXCEEDED the 355 Mb cached read spec. You will get different answers with different kernels. The drive does weather very well, if one is very gentle. I suppose this method is superior, because I think that simplicity is the best asset. Also make sure that your fstab reflects noatime,norelatime,discard in the mount options. This might not fly in high-security enviornments, though.

Peter Joseph Harris
  • 11
  • 1