No, this isn’t about pork bellies
In just the last few weeks EMC and IBM have announced their intentions to offer commodity server-based storage. EMC with Hulk/Maui and IBM with XIV.
Sun already offers its Thumper product, a high-density server and storage chassis with 48 disk drives, at prices competitive with commodity servers from a GB perspective. You can put OpenSolaris on anything you choose.
A few more shoes will be dropping
So EMC, IBM and Sun have commodity-hardware based storage. HDS, HP LSI and NetApp don’t. How long will they choose to hold out?
HP is part way there. Polyserve runs on HPs popular blade servers. But if density isn’t your goal and scalability is important, you may lean to another solution.
Pressure on the margin
It is a simple question, really. How will expensive hardware compete with cheap hardware? After the FUD and hand waving are said and done, customers of enterprise storage will have a new choice with a lower cost structure.
If HDS and NetApp aren’t in the new game, where exactly will they be?
The StorageMojo take
HDS is not nimble, so I’d be pleasantly surprised if they did anything in 08. NetApp also has some breathing room as none of the big boys are offering commodity cluster NAS. But that could change in a week.
NetApp appears to be the most vulnerable. Their largest customers are ripe for conversion to a more scalable architecture and lower costs. No matter how much NetApp discounts, their costs are higher than commodity hardware. They can fight for a while, but not forever. They have to be competitive and their big customers have to believe they will be competitive.
Expect to see NetApp make a cluster storage software acquisition in 2008.
Comments welcome. Is Polyserve going to meet HP’s needs going forward, or are they too going to have to buy or create a new cluster storage product?
Until EMC gives more detail on their Hulk/Maui I’m not going to get too crazed about it until more details other than “some kind of commodity storage cloud is coming” as anticipated expectations relative to reality can be dramatically different things (but I can say that I’m drooling for some player to release something soon).
I am a happy NetApp customer, but I do agree that NetApp is probably going to have a large difficulty in dealing with bulk storage needs. NetApp’s sales pitch to the world for the past decade is: a single platform, a single architecture, a single stack. They’ve been trying for years to get GX out the door and into their singlestack, they finally have some very restricted fragments into it; but it is pitched to be something completely different than bulk storage (HPC which is pretty much the polar opposite). For them to try and push something else into it in any quick timeframe will be about impossible, so they will have to purchase someone to have any tranction in this space. If they purchase it will be very interesting to see what they do about their main sales pitch, will it be a single platform, and we just won’t admit that bulk storage thing over there isn’t part of it’s stack (their VTL is not running OnTAP) and still make that our pitch.
I really like the idea and cost of the Sun Thumpers but I don’t like their management size, in that I have 24TB and the box is maxed out, I don’t want to have to manage 20 little islands of storage. Hopefully the marriage of ZFS & Lustre can help with those islands, but I wonder if with the acquisition of mysql, will mean that Sun has to focus elsewhere for a while (screwing up a billion dollar acquisition would be suicidal) and it’ll be sometime in 2010/11 when we will finally see the merge.
To me commodity lends itself to massive quantity explosions, just like when we made the transitions from big-iron to little pizza boxes because we now could do it at a lower-cost point. All the things businesses have been wanting to do but couldn’t they are going to want to do now (and end users will get even lazier about keeping up their storage clean since it’s now so cheap). That’s why I say the words bulk, as I think that’s ultimately what businesses are primarily interested in from commodity storage.
There’s a number of interesting small players in “cloud storage” that are pretty much running under the radar; I believe they will become on the radar very soon. Unfortunately they are so under the radar they will need to be legitimized before businesses pony up for a petabyte of storage using them, but an acquisition often does that.
Robin,
HP is almost their with its coming MSO product – http://www.techworld.com/opsys/features/index.cfm?featureID=3932&pagtype=samecatsamechan
NetApp – yes – we’re waiting for something to drop. HDS could partner … Isilon?
Clustering may not be the only option. NetApp could probably replace their controller with a commodity box running ONTAP and then attach disks using SAS JBODs; the result would probably be even cheaper than cluster storage, although not as scalable.
And isn’t StoreVault already commodity? Maybe they will simply allow larger StoreVault systems to be built.
I’m not certain what you mean by commodity here:
NetApp products, with the exception of software and their NVRAM cards, are primarily based on off the shelf or commodity components (AMD processors, high end ECC RAM, etc.). I would state that they were the first big “commodity” storage player out there, at least in how their gear is constructed.
Using polyserve as an example is also interesting, seeing as they require a substantial existing or new FC SAN fabric to run, which typically requires a classic storage array of some sort. (NetApp, HP, EMC, etc. etc. etc.) Polyserve is not a Lustre like DFS or CFS.
Isilon would be a better comparison, and they haven’t been doing so great lately.
Also forgot to mention that NetApp already made a cluster storage software acquisition… I believe the purchased spinnaker some time ago and sell that product under their GX line.
i understand it is not a commodity storage product, but if NetApp puts the time into validating other storage vendors on their GX line, will that not give them some advantage? granted the GX line requires “ownership” of the storage arrays behind them, but still if the right arrays are validated and supported, NetApp could steal the JBODs of the data centre. just speculation, please correct me if i am off the mark.
tim
Exactly what sort of cluster storage software were you expecting them to acquire? They already have GX…
Robin,
I’m surprised you didn’t mention Isilon : commodity hardware, true cluster architecture, etc. An excellent proposition as far as I’m concerned.
Anyway, I agree with your view on NetApp : they’ve completely lost their edge and don’t understand storage needs for digital media. Which is where MASSIVE growth is occuring right now….
Commodity hardware + pNFS + ZFS/Lustre/OneFS : this IS the new breed 🙂
AFAIK GX is far from being ready for production, though they happily market it. The few people that actually use it are under some sort of beta test actually.
GX is very well production-ready and fully supported (we have been running an 8-node cluster since Mid-2006). Of course it lacks many features compared to Ontap 7 but what is there works reliably.
The nice thing about GX architecture is that it does *not* depend on features of a particular interconnect (like Infiniband) or a global, shared cache (which can reduce reliability). With pairs of failover-enabled nodes as basic building blocks there is choice (depending on requirements) between keeping data on exactly one node (using a “normal” volume like in Ontap 7) and distributing it to several or even all nodes.
I see one major issue with Isilon
Isilon thinks they are the next NetApp. They are building a significant amount of tools that only run on their platform. ILM, Snapshots, Replication, TCP Optimzation, Quotas, Reporting..and the list goes on. I don’t want to trade one proprietary technology for another.
Polyserve has been mentioned on here a few times…almost with a negative conotation. Look out for their new filesystem in 2008, it could change the game and make them a real player. HP already has the DL320 which is essentially an Isilon node. Combine that with Polyserve…and you might have some competition. Also combine the fact that you can run Polyserve on any commodity server or storage without worrying about rules for mixing different sized nodes…etc.. Thumper? Hulk? They could all be nodes for Polyserve or some other clustered vendor without a hardware platform.
Why is IBRIX never mentioned in this discussion? They have similar deployments with the DL320 and Rackable hardware as the nodes. Their feature-set is not as robust, but this bulk storage and not standard home directory storage we are talking about.
A real clustered storage system by definition must support linear capacity AND performance scalability with the same reliability and management simplicity as a monolithic system. Globally coherent cache, a robust cluster interconnect (ie IB), and full arsenal of management features are a prerequisite. That is the difference between GX pairing of monolithic bricks which do not meet those standards and Isilon which is a true fully baked clustered storage system. Hybrid systems like IBRIX and Polyserve require cross vendor expertise and support. What’s wrong with a plug-n-store architecture. Simply plug a node to the network and there you have 2-24TB available for use, fully protected, and self-balanced in 60 seconds. Compare that with RAID, volume, file system provisioning and management complexity. When you have a system that is so easy to use and scale you don’t need to worry about capacity planning. When the need arises you simply add storage – now you’re really talking about storage as a commodity which is where this discussion began.
True Cluster Fan,
you sound like an Isilon commercial. I know Isilon’s product as well as anyone. They have a great product as long as you only want to buy their hardware and you want to scale within the confines of a single cluster. Leveraging their new nodes in an existing cluster is painful and restrictive. True clustering allows you to take advantage of dissimilar hardware without such an enormous penalty. If quad core chips are available, you should be able to leverage them. If 2TB drives come out you should be able to leverage them as well. With Isilon’s model you can leverage 6 months later as long as you add 4 and 5 at a time which is a huge restriction.
I am not saying there is a perfect solution, but I want an open solution that can take advantage of hetereogenous hardware. Trading one proprietary venor for another is not the solution.
Lot of interesting comments here. Let me respond to some.
Insane – I look at the impact of Hulk/Maui from a marketing perspective. When EMC endorses a cluster architecture, a cluster architecture suddenly becomes a lot more respectable for enterprise customers. Presumably it will offer storage for a lower $/GB as well, which will eventually percolate into the CFO’s consciousness.
Storage at 20% of today’s cost will change how people use/waste storage. A lot of other players will be affected as well.
Chris,
The HP guy clearly got the lessons of the Google power paper. Whether blades are the answer remains to be seen.
Wes, Max, Tim, TimC,
NetApp certainly has options and they have some time to figure them out. Multi-billion dollar companies take a long time to die unless they have help. NetApp isn’t alone in building a system that overshoots customer requirements. As long as no one else responded they had no worries. Now that EMC, Sun and IBM are acting, everyone has to respond.
NetApp has a least 2 virtualization products, ONTAP GX and the V-series. GX is positioned for HPC while the V-series virtualizes both NetApp and competitor storage. It looks like a standard FAS head.
While ONTAP GX provides a global name space, the backing store is NetApp filers and arrays that aren’t cost competitive with commodity-based cluster storage. If you need a global name space AND high performance, you should be happy. Otherwise, not so much.
ClusterDude – back to work!
I prefer a big-tent definition of clusters. VAXclusters, Google clusters, Polyserve clusters, Isilon clusters are all clusters to me. I’m interconnect agnostic and open to a variety of file locking schemes. The important thing is that we learn how to build upgradeable in place storage pools for long term preservation of digital data.
Cluster Fan – complete solution vs roll-your-own? I think there is room for both.
Cluster Dude Said:
“I’m surprised you didn’t mention Isilon : commodity hardware, true cluster architecture, etc. An excellent proposition as far as I’m concerned.
Anyway, I agree with your view on NetApp : they’ve completely lost their edge and don’t understand storage needs for digital media. Which is where MASSIVE growth is occuring right now….”
Isilon is nice, but it’s less flexible than NTAP products. Great for throughput with large data sets, but aren’t really all that good for transactional stuff. You still need to buy other storage kit for that. If you doubt me, ask MySpace. You can hardly claim to be Unified or a “true” cluster storage system if I have to go by disk from someone else to handle databases.
As for NTAP not understanding digital media, I would tell that to Laika, Industrial Light and Magic, and the rest of the studios who use them in production.
ON the Topic of Polyserve, while the file system is ok (Still doesn’t support quotas last time i checked though ) it’s interface to other systems for NAS is still the same awful ones that the host OS uses. Windows CIFS and Linux NFS. And Linux was never really good as a NFS server; I once replaced a two node polyserve cluster that had awesome hardware ( dual xeons, 4 GB of ram each with 2Gb/s FC SAN with a 270c (600MHZ mips processors and anemic amounts of memory) that for some reason ended up being twice as fast and orders of magnitude more stable than the polyserve cluster.
The issues were with Linux, not Polyserve, but it basically boils down to the fact that the whole issue of vendors using “commodity” hardware is a red herring. They *all* use commodity hardware 🙂
The real leverage will come from commodity software, something that doesn’t really exist yet for HPC / Grid / Cloud or whatever computing is “Teh Future!!!” NFSv4.1 might show some potential with a standards based pNFS implementation, but no one I know actually uses NFSv4 yet 🙂
So, while using a commodity Open Source OS is a good way to get better driver support for the latest and greatest, but it doesn’t actually seem to yield any leverage when it comes to features.
I believe most of the issues with NTAP are pricing related; They’re turning into the new top dog and the arrogance is beginning to show in their sales force. Much like in the good old days of sun, this will piss off some existing customers… but this has little to do with the costs associated with building the products.
I would like to contrast this discussion with another sector of the technology space, I would state that the market that’s low hanging fruit would be your typical software based modular access routers in the data networking space; but last time I checked cisco was still in business and doing fine.
~Max
I have to agree with the others, Robin. You seem to be consistenly railing on NTAP for not having a commodity storage cluster solution, when as someone else pointed out- they do. Their entire lineup is commodity hardware with the exception of a few bits of ‘special sauce (NVRAM card, ONTAP).” Jay Kidd from NTAP is promising the GX/7G merge in 08 (I personally don’t think it will be that quick), and then NetApp and XIV will be the only ‘storage clusters’ that will support block protocols. Isilon, Hulk/Maui, etc. do not.
I believe that clustered storage is the long-term future of enterprise storage, but it must support all the major protocols. Isilon is a perfect example of clustered storage that is falling on its face because their target market is too small. Yes, there are the NFL films and Kodaks of the world, but most enterprises are turned off by their pitch because unstructured (file) data is unimportant to them. Once IBM starts shipping an XIV based product and NetApp moves to the GX platform entirely (bringing all the 7G functionality with it) I think they will really be the only two vendors in the short term offering clustered storage over block.
Re: “commodity” hardware.
You all make a good point: everything is built from commodity hardware these days. But it isn’t all sold like commodity hardware.
Most storage today is essentially “tin-wrapped” software on almost-commodity hardware, with 60% gross margins.
Real commodity servers get sold for 20% – or less – GM. The software is another buy that gets made on its merits and, of course, the GM is huge, as is the SG&A.
To be more precise, some vendors are unbundling HW from SW and some don’t seem ready to make that leap. I’m concerned about how the latter will fare.
TylerB – I don’t agree that a storage cluster needs to support all protocols, any more than a commodity server needs to run MVS. The NAS market has been growing faster than the block market for a long time and it will continue to. It just makes more sense to store files as files.
Block storage and big-iron arrays aren’t going away any more than MVS has. It is just that the growth is in file storage, not block.
As for Isilon, I don’t know where people get the idea the company is failing or has a tiny target market. They haven’t reported results for 2 quarters due to an independent audit requested by the new CEO and their stock has been hammered for missing targets, but at last report their growth rate was excellent.
Unstructured data is a huge market. IDC is drawing a distinction I like between “performance optimized” and “capacity optimized” storage. Big iron storage, including NetApp’s, can handle pretty much anything that gets thrown at them, for a price. If capacity is more important than performance you can save a lot of money going with a mini-van instead of a Ferrari.
Robin
Robin,
I can’t completely disagree (heck even IBM’s Shark is a set of p series servers) with your premise as commodity at the lowest level (white box servers or what have you). I also agree with NAS being the future as well (lower cost, better standards interop) especially with VMWare and Oracle on NFS gaining traction, but the reason most of these ‘cloud’ and ‘grid’ concepts are a massive failure in the enterprise is they neglect the transition phase support. If I’m a IT Director and I bring in a clustered storage solution, I want to be able to utilize it completely with my current equipment/applications/protocols and not have to re architect my environment to take advantage of the new technology.
RE: “…they neglect the transition phase support”
“Smart Storage” wouldn’t care.
It would adapt itself to perform in the current environment.
In the automobile world, which is selling best and making the highest profit margins for the manufacturer?
1) entry level “bare-bones” models
2) “Smart” models that can adapt to a drivers profile, road conditions, driver emergency needs, maintenance, etc.
Which would we like to have? BOTH !!!!!
“Smart Storage” adapts its feature/function set to fit the requirements of the Storage environment. Sort of an enhanced or reduced feature/function set.
You pay for what you use. And you pay it forward.
Marketing and Sales should love this if they were clever enough to figure it out.
The cost of the intelligence to enable driver profiling and the cost of the mechanics to adjust the driver’s seat, mirrors, A/C settings, radio, cell phone, TV, etc. are not the show-stoppers.
Lack of creativity and ingenuity are, and not at the engineering level.
Where is the leadership?
Like Scrooge McDuck they are swimming in their money.
The same is true for Storage…
Why do people talk about Isilon like it’s a dead duck? They just upgraded hw and fit their sw to play nice with vmware. Last I heard they’re still gaining traction. Demand creates demand. File-based market is much smaller, but growing much faster. They’re ‘niche’ might be bigger than people think.