Just a quick post here to point out an interesting StorageMojo.com blog entry about Jonathan Schwartz’ blog entry announcing the server and storage group merge at Sun. It seems Robin Harris thinks it is too little, too late. I think it is a no-brainer and looks like a follow-the-leader move. Why would readers of my blog care?
Monolithic Storage Arrays Must Eventually Die. Good Medicine for DBAs.
Folks, I’ve pointed out through my Manly Man SAN series that grid computing and SANs really don’t mix optimally. Perhaps the point is best made in my post about connectivity and provisioning in a Grid environment. Those refrigerator-sized storage arrays started out in life as very specialized hardware but over time they have evolved into solutions increasingly based upon general purpose processors. Some of them have NT embedded in them, some have AIX embedded in them and so on. The ones that are 100% proprietary will not be able to command those margin$ for ever. What Jonathan is doing is a lot like what HP did by introducing storage servers, and more interestingly, storage blades. Those Proliant guys really got it right. Jonathan wants to leverage the power of commodity computing with a specialized storage twist implemented in software (with a Solaris foundation).
Why Would a DBA Care?
Storage solutions that are more software-oriented bring the database and storage more into the control of the DBA which reduces organizational complexity. That increases the value proposition a DBA has in an IT shop. Such a model equips the DBA to have more control of the total solution (server and storage) since the architecture you should envision is one where the storage servers and databases servers are physically adjacent and both the storage operating system and database server operating system are potentially (preferably) the same. Since Oracle Database 11g offers SecureFiles, you might even consider “internalizing” solutions that include both structured and unstructured data all in one rack-eliminating any chance for cross-organizational strife. Just you, a Sys Admin and a few racks full of very powerful commodity-based servers/blades and storage servers/blades.
That sounds like the good old days when an Oracle DBA team and a good System Administration team were the sum total of manpower required for really large Oracle deployments…you know, back in the days before the current Fibre Channel networked storage paradigm.
Summary
I’m not necessarily saying that what comes out of this Sun move will be some great storage option for Oracle databases per se. What I’m pointing out is that the move should make Oracle DBAs look to the future a little differently.
Oh, Come On…
I know, it wouldn’t be one of my typical blog entries if I passed up a classic pot-shot opportunity. I aim not to disappoint. I’ll focus on the following quote from Jonathan’s blog entry (emphasis added by me):
So we’ll still be strongly focused on being a multi-platform storage provider (just as our servers run multiple operating systems, and our operating system runs on every vendor’s servers)[...]
Egad! How’s the uptake for that Solaris port to PowerPC or, uh, Itanium?
…and if I were Jonathan, I’d ask who actually owns an Itanium server? Anyone other than a locked in HP customer? Solaris doesnt need to support dead architectures to be popular, just high volume ones.
Blake,
Right you are. I was tongue-in-cheek. As a side note, Oracle Database 10g was available on PowerPC Linux _before_ Itanium Linux! Think about that!
I can’t wait to get my hands on those Nehalem processors…but then working at Oracle I don’t quite have the luxury of hardware to play with that I have grown accustomed to over the last 18 years with my Oracle focus… boo hoo I guess… when you focus on Oracle at a hardware company you get lots of nice big toys to play with.
I notice you didn’t comment on
“ZFS - soon to be parallelized by Lustre, a recent acquisition from Cluster File Systems).”
Think it will work? I remember you thinking not much of filesystems that were had parallelization retrofitted.
I totally agree with you post, Kevin.
I have never understood, why one should buy a specialized storage system that in the end is nothing more but an appliance and not a totally “new & different piece of hardware”. History of computing has shown that specialized hardware was and probably always will be replaced by software for being better and cheaper. I think time has come to ask why “storage” (or storage-systems, SAN-based storage, fill-in-whatever-storage-technolgy-you-like) is being treated like a “holy cow” and not being treated as any other computing resource such as # cores or cpu frequency that is available: interchangeable and just another standard ordinary computing utility.
“That sounds like the good old days when an Oracle DBA team and a good System Administration team were the sum total of manpower required for really large Oracle deployments…you know, back in the days before the current Fibre Channel networked storage paradigm.”
I think you’re spot-on though — rapid provisioning is going to be a requirement going forward — and that will take more friendly solutions that can be automated and require fewer hands.
Jas,
RE: “ZFS - soon to be parallelized by Lustre, a recent acquisition from Cluster File Systems).”
If they Lustre-ize ZFS it might work for divide-and-conquer HPC workloads. But no, I don’t expect to see a ZFS-based general purpose clustered filesystem just because of Lustre.
I checked out those storage blades and was not that impressed. The choice is between a direct attached blade (I guess only a single compute blade can use it), and a microsoft powered iSCSI blade. Direct attached storage is just not even a consideration in our environment - too slow, inflexible, sprawly, and difficult to manage. The ISCSI is interesting, but why does it have to be MS??????? Sometimes I really hate HP so much for being bill gates camel.
I don’t really generally agree that SAN and grid don’t mix, but I’m not that much a of a fan of the fully distributed grid, at least as currently implemented, anyway. A million 1U pizza boxes with separate power supplies, network connections, and copies of the OS is the mess that is the modern data center. Mostly it is the fault of the software vendors (really almost everything that’s wrong with modern IT can be blamed on the software vendors) that charge so much less for ‘commodity hardware’ for no technical reason reason, but these 1U commodity boxes are incredibly wasteful, and a PITA to manage (so now they try to sell virtualization/server consolidation to get us out of the mess they created for us). And most of this has been driven by microsoft-centric IT.
I think the term SAN has been pigeonholed to only mean fibrechannel over fibre, when really is should mean, block addressable network storage, including iSCSI, and different types of san are appropriate for different things. As you point out fc-saniscsi or nas gateways are not too expensive and versatile.
I LIKE refrigerator sized storage arrays, having a lot of disks working in parallel is the way to get performance. They are so flexible and nice to manage (well, the HP EVA is nice to manage). As you say, it’s the price that gives pause. We pay twice for storage - once for the disk, and then again for the capacity license. This is evil and deceptive, and customers need to call b[edited by moderator]t.
Refrigerator sized servers, would be OK, but the price here continues to be beyond ridiculous, both for hardware and software. There seems to be a happy medium now in bigger than 1U servers that have a lot of power. the 1U box is not by itself evil, it’s the sheer numbers of them, and if a 1U box is powerful enough, we really don’t need that many, right?
Dominic,
Rapid provisioning that can be scripted - another reason big arrays are nice.
Vendors are concentrating on G-WIZ ’storage resource management’ bloatware they think they can sell to managers who don’t directly work with storage. Any storage infrastructure where you have to point and click your way through providing storage should not fly with the folks who actually have to use it.