The trend in the storage market over the last few years has been towards ‘unified’ storage architectures – one box from which any manner of storage can be delivered. The background to this is that some time back, storage was generally acquired as either ‘block-based’: DAS (direct attached) or SAN (storage area network) to support applications or as ‘file-based’: NAS (network attached) to store and share files and items. Combining both arrangements into one box, delivering both file and block storage over multiple protocols (iSCSI, Fibre, FCoE – maybe even FCoTR!) has delivered savings, particularly in terms of management. Often such a box would have different types of hard drive for different workloads – small, fast drives for applications and slower, cheaper higher capacity drives for file and snapshots.
I certainly think this approach has legs – that it will suit many organisations for some considerable time to come. However, I’m just starting to see the signs that for some companies, a different approach might suit. There are two trends that might just take us in a different direction:
Firstly, the need for speed. While storage capacities have increased by an unbelievable amount, the physical speed with which drives can read and write data hasn’t kept pace. Physics gets in the way, here, as hard drives are mechanical devices. I’m guessing that if you built a hard drive faster than 15k RPM, it might well spin itself apart (or at least the MTBF would be shortened). Enter the SSD, or solid-state drive. As this is based on non-mechanical storage, given the right environment, data can be delivered much faster. Databases, VDI, that sort of thing – these workloads drive IOPS like never before and SSD could be a part of a solution. The downside, of course, is that it is expensive and capacities are limited.
The second trend is what EMC are calling Big Data (I love the lack of buzzwords and acronyms in that phrase!). Files, videos, images… everything is getting bigger, and companies are crunching more and more data – and needing to do it quicker and quicker – than ever before. Just look at Apple’s recent purchase of 12 Petabytes of storage, presumably for iTunes. That’s extreme, and not every organisation needs anywhere near that amount of ‘stuff’, but there’s a need in some cases for large amounts of file-based storage, which can scale dramatically with minimal management overhead and high levels of availability. Other scale-out NAS platforms are available, of course, including HP’s IBRIX acquisition from 2009, which is now known as X9000.
I think that based on these two trends, we will soon start to see companies run their applications off SSD storage (or other solid-state devices), and use cost-effective, designed-for-purpose, capacity from scale-out NAS architecture for their ‘stuff’ – all that unstructured data that exists outside of a database – the growth rate of which is far exceeding structured data (databases etc).
Anybody else share this view?
Sam Routledge's Blog 
April 16, 2011 at 4:00 pm |
Hey Sam!
I’m concerned about the vagueness of big data as it’s being tossed around today. I’m frankly never sure what anyone means until 5 minutes into the conversation. There are 10,000 different meanings being discussed by 100 different people at 100 different (mostly vendor) organizations. That kind of discussion quickly creates confusion and that’s when you start to hear “marketing hype”.
For me, I see the term big data breaking into two core discussions:
1) Big data analytics – take for example the recent acquisition by HP (and I know you know I work for HP but want to make sure your readers know too) of Vertica. I recently saw a demo of a Vertica appliance and what it can do in analyzing large amounts of data quickly is pretty cool.
2). Big data content repository (I don’t like the name I’ve used but don’t have a better one) – and this is more what you described above with the HP X9000 – a scale-out, clustered NAS solution for holding large amounts of data. As an example, the X9000 can manage 16PB of data in a single namespace.
I really hope we get past the hype of the term (and you’re probably right – EMC created it as they have with many over-hyped terms before) and get to deeper, meaningful definitions so that we can all talk about the same thing.
Great questions to be asking in your post!
Calvin (@HPStorageGuy)
April 17, 2011 at 8:58 am |
Thanks Calvin. Appreciate the response – and I agree with you that it is about ‘processing’ and ‘storage’.
Funnily enough, my overall point was that the ‘one size fits all’ unified storage trend will not be right for everyone (although it does have its place) – something I felt was the major flow of Tom Joyce’s recent post at http://h30507.www3.hp.com/t5/HP-Storage-Blog/Why-settle-for-bad-Vanilla/ba-p/90943
Thanks again!
Sam.
April 17, 2011 at 2:03 pm |
Sam
I can only share your view. The capacity of the SSD s is increasing, the price is going down. Since SSD are about 100 faster, it is just a matter of time that the transistions will migrate to SSD. It is starting already with VDI.
The next trend we see is moving databases towards SSD.
Imagine howmany drives customers have today in RAID 1+0 for performance. It is smarter to move these drives in RAID 5 for capacity and use SSD storage for real performance, both IOPS and low latency.
April 18, 2011 at 7:18 am |
Good read.
I don’t allways get it why we would want to have block&file + all protocolls in one box. Why would you have an iSCSI & FC network in the same rack? But being capable of doing all of them with 1 product is nice to have. Therefore I like working with the concept of the HP P2000 (fka MSA). You know the product, just chose your controller and off you go. SAS? No problem, FC? iSCSI? no sweat. Would love to see that concept more on the bigger solutions too (and no, not a ‘proxy/router’ or something like that )
And then there’s VDI. One of the most underestimated solutions for storage. It demands very high performance even for small environments. Adding loads of SSDs to the SAN may help but maybe it’s not allways more and bigger we need. I was amazed last month seeing that at Microsoft MMS the complete environment of that seminar (with thousands of machines and labs a day) ran on just 2 racks with 2 ‘small’ EVA4400s. How did they do that? Offloading the IOPs to the blades instead of the SAN: IO Accelerator. Certainly worth a look. http://bit.ly/48rI4
April 18, 2011 at 7:56 am |
Hi,
Those IO Accelerator cards look very cool. At VMware Forum they also showed off these and I wrote a snippet about it here:
http://www.core-four.info/2011/04/fusion-io-technology-that-allows.html
Computerworld also made a video about these cards here: http://blogs.computerworld.com/18120/what_4_500_video_streams_look_like
Lars