Take the 2-minute tour ×
Server Fault is a question and answer site for professional system and network administrators. It's 100% free, no registration required.

I am in market for a new server and I require heavy disk iops. Our retailer is offering an IBM BLADE HS22 with dual Xeons, but limited to RAID0/1. For some reason the only SSDs he can offer me are 50GB drives for 1000$ each. I am considering buying 2 Crucial C300 256GB and setting up a RAID1 configuration.
So: Viability, pros, cons, should I, should I not?
Keep in mind we need this in a rush, processing power is OK, but lots of random disk IO, especially DB lookups. 256GB is enough for the foreseeable future, but just barely, so no option for Intel's 160GB.

Any suggestions and alternatives are very welcome. I am a lead software engineer by profession, and not by any means new to computers in general, but unfortunately I am new to configuring servers.

share|improve this question
    
Wait - your are buying a blade to install consumer grade SSDs? –  pfo Apr 12 '11 at 20:07

5 Answers 5

up vote 3 down vote accepted

This question seems a bit off. I mean no offense in any way, but are you aware that blade servers require a proprietary blade enclosure and other infrastructure to be present?

If you're using blades, I would expect you to also be using a Fibre Channel SAN or similar for your storage. If so, then you should probably stay with your SAN for your storage needs. If you're not using blades in a power-dense datacenter (most likely together with a SAN), why are you picking a blade server -- regular rack servers are generally cheaper for very small deployments, and have more storage options.

If you want to use an SSD, you absolutely can. Big +1 to Littleman's suggestion to use an enterprise-grade SSD which is designed and built for heavy I/O -- and not a desktop SSD which are typically built for <10GB writes per day. It would seem strange to purchase an (expensive) IBM blade server, and then put a consumer SSD in it (which BTW isn't certified by IBM).

Good choices in enterprise SSDs include:

share|improve this answer
    
You raise good questions. Like I said, this isn't my area of expertise, but unfortunately has fallen to me. Blades is what we use and it sort of came by default. I'll reconsider our options. –  ssteinberg Apr 12 '11 at 21:11

If you're buying a production server and you really need the performance, I would try and get an enterprise level SSD which means SLC NAND flash. The Crucial drive you mention is a consumer drive using MLC flash. A quick search shows that Micron has the RealSSD P300 in a 200GB version. Seagate makes an SLC enterprise drive called the Pulsar in a 200GB version (Anandtech's review). SLC is more reliable than MLC and will last much longer (it's also usually faster). So if your usage profile is heavy on the IOPS, I'd go for SLC. It still might run you a couple thousand dollars per drive. If you're looking to save money, I'd skip the RAID1 for now and see if you really need that extra read performance in practice. Also, keep in mind that the 200GB drives I mentioned both have 256GB of flash but reserve some for "spare area for cleaning/garbage collection/bad block allocation" (quote from Anand) so you may not get all 256GB from the Crucial drive (bit-tech's review of that drive says 17.8GB is reserved for those purposes).

Bottom line: I would say not. Try to get an SLC drive. Really though I don't know your limiting factor. Is it budget? Is it capacity? Is it performance? What's most important? The answer could change based on your priorities.

share|improve this answer
    
Budget isn't a problem. We need RAID1 to keep the server up virtually 100%. Performance of modern 2nd-gen SSDs are good enough. –  ssteinberg Apr 12 '11 at 21:10
    
@ssteinberg Then I'd say just grab two (or more) of whatever SLC drive has the capacity you want. Jesper listed a couple good ones if you don't like the ones I found via my quick Google search. –  Littleman Apr 13 '11 at 22:49

Those consumer-grade drives will last a consumer longer than 3 years. However, you're looking for heavy I/O Ops, so your usage patterns will not be that lenient. Yes, that Crucial gives you a lot of space and a lot of I/O operations, but hammering those drives that hard mean you'll be replacing them a lot faster than 'Enterprise' grade drives. Worse, since you're mirroring them they'll wear very similarly and will therefore fail pretty close to each other.

There is a solid reason why 'enterprise' SSDs cost as much as they do, and that's because they're designed from the chips out to deal with the kind of I/O loads databases and suchlike can throw at them. The top tier of drives use SLC flash, instead of MLC, which is overall faster and lasts longer, at the price of being a lot more expensive.


The one area I'd consider consumer-grade SSDs would be if this database is write-lite for some reason (updates are a small fraction of overall disk operations). Consumer-grade MLC should keep up with that kind of workload. Only you know your read/write percentages, though.

share|improve this answer

We bought some Crucial C300s to test the waters of SSDs. (After all, if the consumer gear kicked butt, then we would invest in enterprise SSDs.) We had two problems with the cheaper gear:

  • The SSDs were mounted in HP DL380s, with the P812/1GB RAID controller. The newer DL380 firmware doesn't recognize that the C300 reports a temperature of 0xFF and so ran the fan on high, then gave up because of overtemp and refused to boot. Dropping the BIOS back a couple of versions 'resolved' that issue.

  • We lost a database because of a power failure -- the drives lied to the system regarding having actually written the data to non-volatile store. When we pulled the plug, the database ended up in an inconsistent state.

The cheaper C300 SSDs were great for testing our app, but I would never deploy with them.

share|improve this answer
    
I am kinda dropping this and going with RAID5/6 with 10K SCSIs. However, I am using a C300 in my personal laptop. Those kinds of problems were common to oldest firmwares. With the 006 firmware I had 0 problems over the last 3 month. –  ssteinberg Apr 14 '11 at 16:48

There are many pros in using SSDs.

I only see two cons.

1- Slower write compared to server grade HDD. 2- Cost/GB

If you are ready to spend the extra money I would go with SSDs.

share|improve this answer
1  
Slower write? Really? –  EEAA Apr 12 '11 at 19:33
    
In a RAID1 the C300 won't be writing random slower then any SCSIs. What I am more interested is in the technicalities, like how will the IBM server behave with 3rd party drives (I know some server BIOSes complain or something, again no experience here), how easy will it be to set this up, etc. –  ssteinberg Apr 12 '11 at 19:36

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

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