3 Comments

Summary:

With its Project Lightning server-side flash cache, aka VFcache, EMC hopes to show itself as a forward-looking storage provider. But until it loses its big box, scale-up mentality, it won’t be much of a factor in webscale data centers that go for scale-out everything.

3762193048_ecaa18a3a1_z

With its Project Lightning server-side flash cache, EMC hopes to show itself as a forward-looking storage provider. EMC’s problem is that it, unlike newer competitors such as Fusion-IO, it is not focusing exclusively on the types of storage that sell into webscale data centers because it must protect sales of its legacy products. The bad news for EMC is webscale is where the growth is.

Massive webscale data centers use commodity hardware.  The Open Compute Project, born of Facebook, but embraced by Rackspace, Intel, even Goldman-Sachs, specifies standard architectures for servers and eventually storage, targeting these huge, webscale data centers. Companies like EMC with a lot of proprietary IP to protect, do not fit into that vision very well.

As more workloads move into these big Amazon- or Rackspace-style public cloud platforms, they are lost to these old-line IT powers — the ranks of which also include Cisco, Hewlett-Packard and Oracle.

These legacy hardware companies are pursuing a scale-up strategy–vertically integrating more of their own compute power, storage, networking into single, pricey boxes, as opposed to the scale-out boxes underlying public clouds. In scale-out mode, huge numbers of inexpensive server are strung together to divvy up tasks. Adding more boxes, adds more scale.

Legacy vendors on wrong side of scale-out divide

The pitch from the likes of EMC or Oracle is that vendor integration holds value that customers will pay more for. Oracle even characterized Exadata at its launch a few years ago as a “cloud in a box” and Oracle CEO Larry Ellison claimed that a pair of Exadata machines could run all of Facebook. The problem is that there’s only one Facebook and it’s not running Exadatas. There is a market for those boxes, it’s just not growing as fast as the webscale world.

GigaOM Pro analyst George Gilbert summed it up: “The big disruption for EMC is that they’ve built their business around being the mainframe of the storage market. If you look at the webscale data centers, they’re all about scaling out vs. scaling up.”

EMC — like Oracle — is also hamstrung by a business model predicated on the sale of very profitable big boxes. The company fields an aggressive sales force that gets paid commission on these things. Needless to say that model won’t fly in the webscale world where Amazon or Rackspace buy thousands of servers at a time and can pretty much dictate their terms because their buying power is so huge.

Both Amazon and Rackspace wring every extra cent of cost out of their infrastructure which they then rent to high-priced software developers who build software and services for business users. Oracle and EMC, on the other hand,  sell high-priced gear to companies that then use lower-priced developers to tweak software and services. The difference to tech buyers is they are trading capital expenditure (capex) spent on Oracle and EMC for operating expenditure (opex) that flows to Amazon or Rackspace.

No one thinks EMC is going the way of the dodo bird, but until it and others in the scale-up world get the memo, their growth in these new data centers will be stunted, to say the least.

Photo courtesy of Flickr user redjar

    1. i was there. remember it well. thanks for the message

      Share
  1. What about products like Isilon and Atmos? They are of the “scale out” designs that you are speaking of. The other thing is that the majority of sales still come in the mid range of the market with the CX line and now the VNX series. The sales of these machines are some where near 4 to 1 compared to a Symmetrix. BTW, they don’t support “frame” either, if you want to hook up to your IBM with Ficon/Escon you’d better be getting a Symm, IBM or Hitatchi array. The thing to keep in mind is that I deal with machines in the petabyte range routinely, but not everyone needs that kind of capacity. I mean why worry if your gear is “web scale” when you need say 50TB. There’s a lot to be said for hooking up the power and fiber channel, running some commands and things “just working.” Just like things such as Mongo DB don’t make sense when you’re dealing with structured data in the gigabyte range and MySQL does everything you need for it to fine without the learning curve, the configuration headaches and having to code your applications to work with it.

    Besides that EMC is also doing HDFS for Hadoop on Isilon as well as supporting it via Greenplum’s later DB releases. Hadoop is probably going to be a huge market in the next few years, and they are positioning to be in a place where enterprise people can have turn key capabilities when they need it.

    No, I’m not arrogant enough to be like Old Man Larry there and say that you can run Facebook on 2 racks of Greenplum, but as of now you can get a configuration with about 4.5TB RAM, 1152 CPU cores and 1152 drives. It’s not even the biggest one I’ve heard of (you can get an RPQ past 6 racks.) That being said, the idea of how much raw compute/storage is there is flat out mind boggling. It is also a “scale out” design as well since I forgot to mention that.

    Yeah, it’s a different world out there but EMC also owns VMware which plays a huge role in many people’s enterprise cloud environments. Look up VCE as well. These ties with Cisco make for an obvious channel partner in sales of this equipment besides other people we do business with heavily as of now.

    BTW I work for EMC, just as a matter of disclosure. There’s a lot of facets to what you’re talking about, but the world isn’t as closed as you make it out to be by any means. Quite honestly, we love it if people buy our stuff (hell it keeps me in a job) but honestly if you look at our history we play a hell of a lot nicer with others than say Oracle does. Keep in mind that with Greenplum it’s really the first machine that EMC makes that does actual computation, rather than store information for accessing via someone else’s mainframe or servers. We also invest in open source technologies and use them in many products as well. As an example at its foundation Greenplum uses PostgreSQL and has made contributions in the form of code to the community. Recently they also worked with Intel and a few other vendors to build a 1,000 node cluster for Apache to use for Hadoop research.

    Share

Comments have been disabled for this post