Posted by: Taylorallis
Application-Centric Storage, Database, DataManagement, EMC, Exchange, HDS, SAN, Storage Vendors, Virtualization
My partner-in-crime Randy Chalfant has also commented on the report and blogs – but I’ll give you fair warning that my friend’s storage knowledge is only surpassed by his passion – and he holds no punches!
In a nutshell, Andrew discusses the benefits and challenges of SANs. He concludes that SANs have not lived up to their expectations, and a new approach should be evaluated. This new alternative is “Application-Centric Storage” in which he defines as storage infrastructure managed by applications like mail, database, or hypervisor apps. If data management functionality (snapshots, replication, provisioning, de-dup, etc.) lives in the application layer – then you only need commodity disk (JBOD or RAID) on the backend. (This is not unlike Sun’s “Open Storage” message of which I used to blog often on when I was with them.)
What’s wrong with SAN?
The report lists four current-day SAN challenges: On the first issue with SANs (low utilization rates) I give Andrew kudos for because he hits the nail on the head. The second issue he lists (limited workload-sharing) is really not a SAN issue at all. Same with his third issue (vendor heterogeneity) – not a SAN issue, in fact it’s an issue with ANY storage solution. His last point (block storage has limited information context) is also valid.
Poor Utilization: On low SAN utilization rates, a distinction and definition needs to be made clear:
Allocation = How efficiently you use what you buy – allocated storage space vs. how much storage you bought. BUT if you over-allocate your storage, or do not allocate space efficiently, this rate can look good. This can be misleading if not measured the right way – you can overlook a significant amount of wasted space.
Utilization = How frequently data is re-referenced (i.e. utilized). We have found that up to 40% of a primary disk array’s capacity holds data that is inert – data that has not been referenced in 6 months or more. Does anyone want to spend tier 1, disk/SAN prices on data that is not referenced for over 6 months?
Some storage vendors use the terms incorrectly or interchangeably – so if a storage manufacturer says their arrays offer great utilization rates, make sure they are not just talking about allocation. Mr. Reichman gets this right – and his numbers show utilization rates around 20% to 40%. This is absolutely what we see in the industry.
But I disagree on what he says the root cause is – SANs!?!? You can find utilization challenges on DAS and NAS. Randy rants on the root causes here – but lack of good data management practices, good storage technology implementation, and other process issues are the root cause – not SANs.
Limited workload-sharing: I commented on a good point Chuck Hollis makes – the Forrester report mixes technology issues with people issues. The report says SANs are challenged because individual departments silo their storage infrastructure and applications. Storage silos have to do with data management practices (or lack thereof) and little to nothing to do with SAN technology.
The New Solution?
Another kudos for Forrester – they proposed a solution. A pet peeve of mine is when someone critiques something without offering an alternative approach.
The new approach according to Forrester is Application-Centric Storage: Basically, embed data management software in an application and directly attach cheap RAID or JBOD to it. And if you wanted to get creative you could cluster these systems together at the file system level.
The Benefit: One of the largest benefits to this approach is that applications give data context and support business objectives directly – so they are in a better position to manage and tier data (i.e. ILM). I couldn’t agree more.
The Challenge: I don’t think Application-Centric Storage will disrupt the current SAN model anytime soon and here is why:
Storage as a Feature: First of all, Forrester gives a list of storage software that can make its way into applications like Oracle, Exchange, etc. (snapshots, replication, reporting, thin provisioning, deduplication, etc.). While some applications and file systems are starting to offer this functionality, I have a hard time believing they will offer all of these features at a level that storage manufactures do currently. I came from StorageTek’s RD&E department and have seen my fair share of development roadmaps. The first items to work on are the software’s core value and offering. The first items to go are the ones that are “nice to haves” but don’t significantly enhance the core offering. Storage management technology will always be secondary to most app vendors and receive less resource than other application features. This will also hamper storage innovation…
An application vendor didn’t come up with data deduplication, a storage start-up did!
App admins take on storage: A second point made is that application admins can manage storage better than storage admins because they know what the data is being used for. Same problem as above – storage is not their core competency nor do they want it to be (otherwise they’d be storage admins!). One needs to know storage process and technology – even if your storage is directly attached. There is a way to prove my point as well – take all the tasks of an overworked storage admin and give them to an overworked application admin and see what happens…
The solution is not to throw out SANs and their storage administrators with them.
The answer is for application managers and storage managers to work better together. And I put the responsibility on the storage people to do this – storage should support application requirements which support business requirements – so storage departments should be surveying application departments monthly on what their requirements are.
Application-Centric Storage Availability:
So what’s available today? Some approaches mentioned include:
Microsoft Exchange: They have a whitepaper on an Exchange + DAS solution.
Oracle Exadata: They use an application-based data volume manager. This is a great feature, but only good for your BI and data warehouse needs. What’s more, they use a proprietary Infiniband network which doesn’t show much more cost benefit over FC SAN.
VMware: I am getting excited about the new storage features that are popping up here. Another point here was made by a brilliant colleague of mine at Sun:
Several customers have deployed SANs specifically to support growing VMware environments!
The Bottom Line:
Do You Really Need A SAN Anymore? YES.
If you are running Exchange or Oracle ExaData and want storage solely dedicated to these apps, then check out Application-Centric Storage. For everything else, look at networked storage as a viable option.
With that said, I can almost guarantee that the allocation and utilization rates for your storage systems (including SAN and everything else) are not where they need to be. The solution is NOT to throw out your SAN and deploy something else. The answer is best stated by Randy’s blog on this subject:
“The best answer for you is to better manage what you already have against a criterion that actually means something to the business. In other words – not technology for technology’s sake, but infrastructure for business’s sake.”