Dan - Please, give me a break. I was working on TI's VME based computer / control systems 20 years ago - they are nothing like blade architectures. You are revealing a huge lack of knowledge about blade designs with these silly statements that compare blade designs with the various modular designs that have been out over the years.
BTW - did you ever work with Intel's multibus designs, of about the same era as VME? Lots of modular systems were based on that approach. It's not a blade architecture either. I was building triple-modular-redundant systems out of multibus in the mid-80s... you could replace every component in the system including the power supplies and backplane without ever missing a beat, and without ever losing fault tolerance.
I agree that CompactPCI is not an enterprise interconnect - it is also not the interconnect that Intel / Compaq are using in their blade architecture. The end point for blade interconnect is infiniband. The near term implementation is not but it uses interconnects which have many of the same capabilities.
And as to your comment "The only sense in which SUN doesn't have a "Blade Architecture" is that their technology is real and not slide ware." ... Sun has crap clusters but at least they have something - 4 nodes with a weak scheme. I would claim that 96 node VAX clusters are the model that most clustering technology was built on, not SUN's johnny-come-lately joke. Or we might look at Tandem's 4096 node clusters - a true modular architecture which allows dynamic reconfiguration of memory, I/O, storage, and processors dynamically. Architecturally like a blade architecture but physically big, and at a little different price point.
You seem to be confusing modular components in an SMP environment, and clusters, with blade architectures. While there are a few similarities, there are not many. And even in SMP, Compaq had an Intel-based SMP design (the SystemPro, in 1988) 5 years before Sun got around to offering their first SMP machine.
And as to "More and more companies are consolidating applications onto single, large capacity servers - like the 64-way Sun Enterprise 10000 system - to lower costs and ease system management. " - not supported by any data. The number of large machines is actually decreasing. UE10000 has replaced a few mainframes and a few customers have been foolish enough to try server consolidation on those boxes - but a recent Gartner study showed that the cost of the consolidated Sun systems was almost twice the cost of the previous separate systems - hardly what Sun promised.
And the total numbers are not exactly earth-shaking. Sun has sold less than 3,000 UE10000 systems IN THE ENTIRE HISTORY OF THE PRODUCT.
And dynamic partitioning is a cool idea - too bad it doesn't work on Sun gear. You can buy systems that are actually dynamically configurable from Compaq, if you want.
Your whole line of argument is the typical Sun crap - whatever is hot, Sun claims that's what they have and have always had. They're now on the web services bandwagon too -hoping to morph their networking schemes to make it look like they have something. IBM at least is actually doing a layer on J2EE that will support web services. I'm not sure Sun even knows what they are, based on their press so far. |
