Sun looks to break through the cloud

Hamilton: "Fewer and fewer applications need proprietary systems"

Sun Microsystems made a net loss of $209m (£145m) in its last financial quarter, and plans to have shed 6,000 jobs worldwide by the end of this year. UK-specific numbers are not available. Despite this, and rumours of a possible takeover by IBM, the company’s vice president of Americas systems practice, Marc Hamilton, insists that Sun is better placed than rival vendors to weather the current economic turmoil.

Computing: How does Sun compete with larger rivals in such difficult market conditions?

Marc Hamilton: Attracting customers by being a large outsourcing or services company works for IBM and HP, more so since the latter’s EDS acquisition, but not for Sun. Similarly selling servers and storage on the back of laptop or desktop PCs, like Dell and HP, is not for Sun either. Rather, the third approach is to attract customers through open source software, and, theoretically at least, monetise that by driving them to an open set of cloud platforms [Sun is a supporter of the recently announced Open Cloud Manifesto].

How do you make money out of open-source software?

It ties in to cloud computing and being able to go into a company that we know has downloaded and registered so many copies of MySQL, for example. Then we say, "Would you be interested in buying a MySQL support licence to indemnify your ownership, or subscribing to a MySQL backup service?"

OpenOffice documents can be opened in the cloud and saved to the cloud. At that point we can offer things like free 100MB of storage space to the certain percentage of those who register with us (typically around 10 per cent) and provide a username and password. That helps us further identify the users and companies using the software.

There’s also a good chance that anybody downloading the Lustre HPC file system, which currently numbers around 15 per day, will also buy a lot of other products we sell.

With so many other providers doing cloud-based data backup, how do you extend the business?

Once you get beyond basic storage services where the likes of Amazon have been successful, the most natural offerings are vertical clouds tailored to a specific set of applications and/or independent software vendors (ISVs), like automotive crash analysis or fluid dynamic design, for example. This is not provided by general-purpose public cloud, and can provide middleware, databases and other applications to organisations that have not thought about how they look at utility computing models to license software yet.

IDC predicts HPC servers will be one of the few growth areas this year. Where does Sun feel buyers are coming from?

Over the past three to four years, customers realised that there were fewer and fewer applications that needed proprietary systems or mainframes that could not also use HPC clusters for the same purpose. Sun initially missed out on part of that, but the opportunity is not just around faster computers, but also networking and storage.

People still go out and buy expensive proprietary storage: x86-based server clusters, disk drives and RAM, and pay for proprietary software for Raid and backups. We say that all of those storage features can be implemented on HPC servers with open-source operating systems and Flash disk technology, with storage services wrapped in.

Aren’t solid disk drives much more expensive than spinning disk equivalents and therefore less likely to prove attractive in times of shrinking IT budgets?

Even though Flash storage is several times more expensive on a per GB level, it is less expensive on a performance level. You need the [server] OS to have a file system that is aware of Flash storage, which Sun has, otherwise you would have to optimise today’s databases to use a Flash drive, which is a manually intensive operation.

Where does Sun’s activity within cloud computing and HPC leave the grid/utility computing model?

The total need for compute cycles will continue to be driven by available budget, and by organisations in aircraft design, oil and gas discovery, movie production, and the traditional academic and research environments. They may not need to have as many computers in their own datacentres, so it is really an equation around how many bytes of computing are they doing and what is the cost of doing that over the network. The ones more likely to move to a cloud model are those with low network latency and costs, because they can see economies of scale.