I'll have to admit, the perceived promises of Sun's Thumper box were lost on me initially. Likely this was because I saw the form factor and specifications, and immediately dismissed the system as just another large server box. I filed it away on the fringes of memory as so irrelevant that I had to think pretty hard recently to remember what Thumper was. But someone the other day suggested to me that much of the industry is perceiving of the Thumper platform as a trend which will re-invent the industry around system distribution again, steering us away from our consolidation trends.
This hope is placed on a high end server box with a large amount of storage capacity that can flexibly serve a number of different roles with one piece of hardware through the featureset available in Solaris 10 for system isolation and psuedo-virtualization. But the proposal that collapsing storage and server resources into a single commodity unit might further enable the enterprise is a pretty big stretch.
Don't get me wrong, I love Sun's hardware, especially their T1 processor which has absolutely knocked my socks off. And the growth of capabilities in Solaris 10 compared to Solaris 9 is impressive. Bringing the full breadth of their operating system to x86 hardware (I actually run Sol10x86 on my laptop), will likely drive capabilities into the enterprise like we can't anticipate today. There's good stuff going on here.
But Thumper doesn't make sense. There is some application in places, but there has been speculation this platform will revolutionize the enterprise. Hogwash.
Business intelligence and consolidation
One place Thumper is pitched, is in the role of business intelligence appliances, to aggregate and rapidly analyze and search data. But decentralizing data into distributed systems delivers minimal cost savings to the enterprise when you trade off the efficiencies that can be gained from centralized data management, and this holds true in general across the board, even for purely aggregation points for business data. Even with nifty features like failure in place which attempt to commoditize the association of storage capacity with server capacity, you see increasing costs of ownership when you start scattering pods of storage across the enterprise. And you have to be wary of the future with deconsolidation - throw some data out there today which is transient, and tomorrow it might become the enterprise's treasure which has to be protected at all costs. You lose your flexibility to adapt to those changing needs with de-consolidation.
Another place Thumper is perceived to fit is in consolidating systems onto a single platform. But provisioning tools for the management of distributed systems are minimal at present, and I for one can't see a readily flexible architecture emerging from this system that provides for flexibility and transparency across multiple (think lots) of systems, especially once you start isolating storage inside those systems. HP's been chasing grid architectures and flexible provisioning for a while, but even so, everything in this space is pretty immature, and where capabilities do exist, they've frequently been brought to bear through the efforts of 3rd parties. I'd say Sun has less established enterprise oriented groundwork in this area, and with their past isolated control over Solaris, has fewer opportunities to engage 3rd party systems to further their provisioning capabilities.
And finally, while containers and zones are great, Sun's still a little too far away from virtualization to be thinking of coupling increasing data densities with the server layer as a solution set for the enterprise. This is a pitch for a Solaris only shop, and to be truthful, as nice as that sounds in some ways, it's more limiting than most of us can deal with these days. If I'm a Windows shop, and I want to consolidate here, then suddenly I'm back to looking at this unit as a big commodity server and lose lots of the selling points like integrated ZFS. You might iSCSI serve your windows clients, but as an infrastructure manager, I'd rather see a virtualization platform where I can seemlessly manage systems in a large heterogenous pool. Looking at Thumper as a heterogeneous consolidation point just complicates the tangled web.
A solution like Thumper needlessly saddles a flexible infrastructure with storage and power consumption. Sun already takes the cake with building flexibility into the Unix side of the enterprise in their containers, zones, and ZFS capabilities, especially when coupled with a powerful processor layer and flexible storage attachment. Why associate storage so tightly with this layer that we lose the benefits of abstraction?
Where Sun does leverage some value with this approach, is in a killer server platform that can handle a lot of data, and in the right place, might effectively consolidate enough systems to make it the primary enterprise server platform without requiring distribution of data across a bunch of different systems. This might be in mid-sized and smaller enterprises, but you have to ask, amidst data bloat, how long will an architecture here suit your needs? And even where there might be a fit, I'd find myself wondering if it's really worth fragmenting a consistent traditional enterprise infrastructure architecture which separates server and storage.
I for one am hoping this is not a sign of future strategy at Sun. I don't need my storage tied to my server world, it doesn't suit my needs. Don't bet the farm on an approach with such limited versatility Sun, I like you too much as a company.