SSDs Through a Critical Lens

Arthur Cole

Expect to see a growing number of servers with on-board solid state drives (SSDs) this year and next as the demand for both faster throughput and lower cost operation kicks into high gear. HP and Sun are talking up plans for server-mounted SSDs, with other vendors expected to follow suit relatively quickly as the cost of traditional magnetic storage continues to rise.


But questions are being raised as to whether SSDs can live up to their promises, and even if they do, whether loading them directly onto the server is the best way to use them.


While SSDs might not match overall storage capacities of traditional disks, they are certainly superior for high-throughput environments. As the SSD Gurus point out in this blog, even if you stripe the rotating drives in your array to gain the random IOPS performance of an SSD, latency will only build in the server queue to the point where it will impact the transaction integrity of the database or whatever application you are using. SSDs push the IOPS through without a server queue.


But that simply begs the question, as EMC's Chuck Hollis does in this blog, of why you would want to bury such an efficient storage solution in the server? EMC clearly has a bias in favor of SSDs in the array -- although it's still being coy about its plans for the CLARiiON system -- but the point here is a valid one: Wouldn't SSDs be more valuable to the enterprise as a pooled storage solution?


The claim that SSDs are dramatically more power efficient is also coming under increased scrutiny. While this report from Tom's Hardware compares the impact on notebook battery consumption, it's still interesting that SSDs draw more power than 7200 rpm 2.5-inch drives. SSDs still undercut the 15,000 rpm enterprise drives, but perhaps not by as much as some of us have been led to believe.


The march of SSDs into the enterprise is probably inevitable at this point, but that doesn't mean they are the preferred solution for every configuration or application. Before you deploy, it's probably a good idea to figure out how they are to be used, and then crunch the numbers to see if they will really provide all the benefits you hope for.

Subscribe to our Newsletters

Sign up now and get the best business technology insights direct to your inbox.


Add Comment      Leave a comment on this blog post
Jul 8, 2008 5:04 PM Pete Steege Pete Steege  says:
Given the hype, it's amazing that SSDs actually draw more power in a notebook PC than a 7200 rpm disk drive. Just goes to show the importance of testing new technology with a dose of reality. Reply
Jul 11, 2008 6:33 PM Georghios Kypragora Georghios Kypragora  says:
One thing is for sure DASD hasn't kept up with CPU performance improvements especially on the RANDOM READ front. I'm sure a lot of CPU is wasted and a lot of slow performing batch jobs are due to RANDOM reads. Assuming an SSD is available the questions remain:1) How do we perform synchronous replication?2) What is the tranfer speed of the data from the device to memory?Is this why the SSD's are mounted on the Servers? What gain do we get in tranfer speeds? 3) Since RAM is getting cheaper what will come first the SSD's or the Server memory? Isn't the protocol and the tranfer speed a lot faster? 4) Looking at new servers will SUN and HP restrict the Server memory so that they can sell SSDs? 6bit addressing is a fact of life! Reply

Post a comment





(Maximum characters: 1200). You have 1200 characters left.




Subscribe Daily Edge Newsletters

Sign up now and get the best business technology insights direct to your inbox.

Subscribe Daily Edge Newsletters

Sign up now and get the best business technology insights direct to your inbox.