|By Brian Reagan||
|November 24, 2009 04:59 PM EST||
Recently, a blog post appeared on the www.techopsguys.com talking about our technology and our position on SSDs. It’s a good post, and I’m very glad to hear our field team was engaged with the author of the post, but it contains some misperceptions that I’d like to clear up.
First, the blog makes several points about our view of SSD that are based on some recent articles in the trade press. The articles were originally triggered as a result of my speaking at Flash Memory Summit back in July, and they drew the conclusion that Xiotech was integrating SSD into the Emprise line for imminent release. The first part of that is certainly true – we are indeed working on integrating SSD into the Emprise line, but the second part is not. We are not going to release SSD in any imminent fashion, and the news that we have completed integration is premature.
Our take on SSD – SLC flash SSD in particular – has been and continues to be that while SSD is very interesting technology, significant work remains in order to fully leverage the technology. This includes but not limited to:
- Reliability over time, especially in mixed (read/write) enterprise workloads
- Internal algorithms such as predictive erase-ahead, efficient garbage collection and advanced wear-leveling
- Improvements in erase speeds, especially as the device fills or is at capacity
- Page- and block-level rebuilds as wear-out thresholds are reached
- Warranty – the devices must be reliable enough to be warranted for at least five years under continuous maximal write workload, just as HDD is so warranted within our Emprise products
Xiotech can speak with context on this subject, because we became the first vendor to offer SSD in arrays when we first offered DRAM-based SSD in our Magnitude 3D 3000 system in July 2006. In fact, we looked at SLC flash SSD shortly after that release, and found the technology wanting. It’s certainly improved since then, but not quite to the point of full deployment with the same (or better) characteristics as found in the ISE. We think that timeframe is over the next two years, with better technology as time elapses.
A few other items that we’d also like to clarify:
- The post mentions that we are “doing something special with the SCSI protocol” in order to deliver the IOPS we do with our ISE-based storage systems. We certainly perform a number of unique and highly intelligent techniques within the ISE, but what you are referring to is our patented RAGS technique, not SCSI protocol. Amongst many other things, using RAGS enables the Emprise 5000 to get 20-30% more IOPS per drive than normal – i.e. the same drive in an unintelligent bay.
- The post said it would like to see us test a high-end 64-node Emprise 7000 for SPC-1 and show the results, so that the author could better compare large-scale systems with other vendors. The Emprise 7000 does indeed scale very well, as mentioned in the blog, and we have in fact tested a large Emprise 7000 configuration. Those results are not published yet, but we can say with certainty that the results are superior to the array mentioned in the blog in several terms: $/IOP, IOPS/disk and IOPS/controller node, amongst others. Also, the array mentioned in the blog does not “hold the world record,” as it turns out. SPC-1 measures one and only one value metric – $/IOP. In terms of overall IOPS, there are two approaches: scale-up, i.e. build a large monolithic array; or scale-out, i.e. use elements in a grid. Several vendors take the former approach, while Xiotech takes the latter. Emprise 5000 is an element which can be scaled out to meet any workload required, be it a single element or tens of thousands of elements in a large collection. This is why $/IOP is the key metric. It is an accurate representation of what the consumer would have to spend to meet a certain workload. Using the element approach, the $/IOP figure does not change at scale – it remains constant
- To the comment about integration, it is certainly true that Xiotech’s ICON Manager for the Emprise 7000 runs on Windows as an MMC snap-in. However, this is not to say that one cannot integrate in large Linux-only environments – quite the contrary. Any environment which supports Web Services – as does Linux – is perfectly germane. Xiotech offers a complete WSDL with sample code for Java, Python and OpenSLP on Linux (and many other OSes) as well as Windows, PowerShell and MS Office clients such as Excel and Visio. In short, any OS with a Web Services-capable environment is fair game. If you like to code in Perl, by all means go ahead and do so using your favorite Perl-based Web Services toolkit (as can be found on soaplite.com or activestate.com, to cite two sources).
To summarize, there has been no “about-face” at all here at Xiotech with regard to our view of SSD. In fact, we have said and will say again that we are “working on it” and will be ready with it when the technology itself is suitable for the Emprise 5000 and its world-class performance and reliability standard for data storage. It’s certainly around the corner, but not quite here yet. As my friend and industry cohort Marc Farley said about SSDs, “We think they will be an important technology in the future and we are working on integrating them into our products in a way that will allow customers to take advantage of their capabilities.” Couldn’t have said it better myself.
- Cloud People: A Who's Who of Cloud Computing
- Cloud Expo New York Call for Papers Now Open
- Red Hat Named "Platinum Sponsor" of Virtualization Conference & Expo
- Cloud Expo 2011 East To Attract 10,000 Delegates and 200 Exhibitors
- An Introduction to Ant
- Cloud Expo, Inc. Announces Cloud Expo 2011 New York Venue
- The Cloud Bubble: Is Computing Becoming a Utility?
- How To Use Social Media in Your Job Search
- Cloud Computing Bootcamp May 18-19 in Prague, Czech Republic
- Telco Perl Powers Telephony With Linux
- Rackspace Cloud APIs Open Sourced
- This Man Should Be Fired from His Job as a Magazine Editor