1 Comment

Summary:

The great things about open source software stacks is that they’re free and they work. The not-so-great thing is that — like many open source projects — they can be difficult to configure and manage. Luckily, hardware vendors are stepping in to fill the void.

Servers? We don't need no stinkin' servers!

The great things about open source software stacks such as OpenStack and Hadoop is that they’re free and they work. The not-so-great thing is that — like many open source projects — they can be difficult to configure and manage. Luckily, hardware vendors are stepping in to fill the void.

The hardware crew

Dell has been particularly busy trying to take the hardware guesswork out of two of the hottest open source projects around: OpenStack and, on Thursday, Hadoop (technically, Cloudera’s distribution of Hadoop). Essentially, these are hardware configurations specifically designed to optimally run their respective software targets. Accompanying both is Crowbar, Dell’s open source software for deploying the software onto the hardware and then managing the existing system.

The big difference between the two offerings is the price tag. Dell’s OpenStack solution is a reference architecture with optional support and services, while its partnership with Cloudera is a full-on product that includes hardware, software, support and services.

In the OpenStack space, Dell has competition from appliance-peddling startup Nebula, which recently launched. That company, led by former NASA CTO Chris Kemp, has developed its own appliance preloaded with OpenStack software and management tools to run the dozens of commodity servers plugged into it. However, while it’s focused on OpenStack for the time being, Kemp isn’t ruling out developing appliances designed for Hadoop, NoSQL databases and other next-generation software platforms.

Scaling out is hard

Appliances, or even tying specific hardware to specific software, can be a tough sell. But when distributed systems are involved, they might be necessary. Certainly, it’s neither an uncommon, nor an unsuccessful, practice with many proprietary software products.

Teradata , IBM Netezza and Oracle Exadata are three very successful massively parallel database offerings tied to hardware appliances. Dell has gone down this path with other cloud products, too, partnering with Joyent to sell its Dell Cloud Solution for Web Applications. EMC is selling appliances optimized and preloaded with both its Greenplum database and the Enterprise edition of its Hadoop distribution.

The consensus, it seems from the outside, is that managing distributed systems is difficult. It might be easier just to buy an appliance or a specially configured hardware stack and focus on using the application rather than installing and supporting the system.

This could be especially true for open source projects like OpenStack, Hadoop and others that don’t necessarily come with enterprise support. There’s plenty of promise, but it’s a lot of work to learn the software and start using it in production with also having to worry about what hardware architecture will provide the best price, performance, reliability, etc.

Software-plus-hardware success

For companies selling appliances, such as Nebula, or just hardware and optional support, such as Dell, it’s probably not too tough a sell getting customers to pay for something they otherwise could get for free. If they want support, they’re probably buying a commercial version of the open source software anyhow. Why not complete the prudent approach and buy specialized hardware and support, too?

Both Nebula and Dell already claim customers in the pipeline for their offerings, despite the fact that they’re brand new or, in Nebula’s case, not even generally available. I suspect new OpenStack member HP will get in on the preconfigured OpenStack system business, too. It already sells does so for its own cloud software. Perhaps Hadoop also will find its own ecosystem of systems partners outside of Dell.

Then we’ll see how this plays out. Is software enough to get mainstream businesses on board with new technologies, or would they rather just open up their checkbooks to ensure everything from the servers on up is done right? Certainly, it will vary by company, but I predict the software-plus-hardware hook will be a successful business model, if only because scale-out infrastructure isn’t easy.

Image courtesy of Flickr user JohnSeb.

  1. Derrick,
    Great article on one of the hottest topics in IT at the moment – how next-gen big data systems will be architected and delivered.

    OpenStack is going to cause severe disruption in this space, allowing a new wave of “cloud supercomputer” vendors to offer systems that are much faster, much more scalable, easier to manage, easier to use, and cheaper than anything on offer today. These new cloud systems will be deployed both in-house and as the infrastructure layer for next-gen public clouds. Some will be based on the OpenCompute model, some will have huge memory capacities on each node for in-memory analytics, some will have super-fast networking and storage etc. Nebula is one early example of this open standards approach, at the infrastructure layer.

    The game, as you rightly point out, is already moving on to what application solutions are going to run on top of these new OpenStack cloud infrastructure systems. At Cloudscale we’ve developed a cloud supercomputer for in-memory massively parallel analytics. Even running on standard Amazon cloud cluster nodes, it’s already the world’s fastest big data solution. But we’re just getting started. OpenStack is how we deliver this super-fast strategic capability in-house, how we deliver it in a form that is not only incredibly powerful, but also easy to deploy, easy to scale, and easy to use.

    The days of buying systems architectures that were not designed with the simplicity and scalability of the cloud in mind are over. Expect a lot of change as the OpenStack juggernaut gets going.

    Bill

    Share

Comments have been disabled for this post