I have heard SSD's are a bad idea for servers are they? Do they eliminate the requirement for RAID i.e. are they more reliable overall?
The cost isn't an issue here. Can someone provide direction on whether SSD is a good idea for server or does SSD still need to mature for a few years before moving away from 15k SAS HDD?
Or should I just go for it?
Thanks
SSDs can be very useful in servers. As with all storage technology questions, it depends on what you plan on doing with them.
SSDs are supreme when you need to be sure that the latency between I/O operations is as low as it possibly can. Enterprise SSDs, especially devices in the "cost is not an issue" price range, have been around for coming on a decade now. In markets where cost can be an issue, they're still a new technology and the market still hasn't figured out the answer to the question, "how reliable are they compared to rotational media."
First of all, you need to know what kind of I/O patterns your server will create, and how sensitive they will be to storage latencies. If it is highly random and high volume, SSDs will allow you to supply that load with much fewer drives than rotational media. If storage unit count is also important (say, you only have so many rack units and an addon shelf of 24 15K RPM SAS drives won't fit) SSDs can help you stay within performance envelopes.
Second you need to know how much storage you need. SSDs are great for performance but they still barely hold a candle to rotational media for sheer capacity. Things start to change if you need both performance and space. If you need multiple TB, that 24 drive 15K RPM SAS array starts looking really attractive.
Do SSDs eliminate or reduce the need for RAID? No. Drive failures still happen. For high-write applications they can happen faster than for rotational media, especially if you're hoping to get 5 years out of the hardware.
Are SSDs appropriate for the OS volume? They can be. If your swap volume/file is on that drive it can provide some nice improvements in speed. But you still want to at least mirror that volume.
The problem you have with ssd's is that they have a (relatively) fixed liftime in terms of disk writes. We did some experimentation with SSD's with servers that hosted a very disk intensive application. Whilst there was some speed increase over 15k SAS drives, we also discovered that running this application would give us a MTF on these disks of about 50 days!
Now obviously this is an extreme case, but it is worth considering the amount of writing to disk your servers do, and whether the speed increase is worth it.
SSDs are becoming comparable with conventional HDs in terms of bandwidth/buck, but have very limited write capacity. OTOH they offer a huge benefit in terms of reduction in latency and power consumption.
They are certainly not a substitute for RAID.
Where they give the greatest benefits for servers is where you can take advantage of the reduced latency - e.g. in a webserver / streaming server serving up lots of files or as part of a hybrid system (e.g. holding the journals for journalling filesystem where the data is on conventional disks).
The amount of writes is usually the limiting factor. There are solutions which mitigate the impact of this - e.g. jffs2, overlay filesystems but AFAIK there are no hybrid solutions specifically optimized to reduce writes to a SSD.
They're also handy for laptops where battery life is a concern.
This depends on what you intend them to be used for. If you are going to just be replacing the boot drives, or other moderate usage drives, then there really is no benefit.
If you are going be be using them for high I/O operations, you have some choices to make about what SSD's to use. I was just on a project where we deployed some of these - Fusion ioExtreme
We deployed these to be used for some pretty intense file indexing operations.
They can speed up/increase database caching tremendously, but the high-end ones (NAND) costs alot and as Sam says - wear down faster than traditional disks.
