Let’s be clear: HP needed to step out from the pack and stand up as a leader. At its size and scale, it was being buried under attacks from virtually every other technology vendor in its space because it appeared to be the most vulnerable. HP’s only choice was to step up and attempt to turn the market or step out of it because it was overmatched.
Well, with Project Moonshot, it stepped up and what is interesting is that the processor vendors ranging from Intel to the ARM consortium are lining up behind HP to play in this new game. It means a massive change to servers -- much like what happened when RAID became viable -- capabilities are going to change massively and prices should start to drop sharply.
This isn’t going to be a market for the timid or slow moving, and while this new class of servers won’t replace everything, anymore than RAID took out all other forms of storage, it will have an impact on the market similar to when Intel drove x86 into this space.
Mainframes were thought to be dead in the 1980s but the work that they uniquely could do has turned them into IBM’s most profitable system nearly four decades after that company agreed they were obsolete. Mainframes continue for two reasons: A ton of mission-critical legacy code requires them and their unique ability to handle massive numbers of queries at the same time, massive I/O, and allows them to stand out as unique in the market.
However, with some types of analytics, and in the still-emerging world of cloud services, you need a combination of massive I/O and massive flexibility. You have to be able to not only handle the mainframe-like level of requests, but because each request may require a completely different level of service, application type and even platform, the systems they attach to are increasingly virtualized. This means that the server has to be able to automatically configure itself around the user’s unique needs and still keep the job isolated and secure from other users’ projects.
Eventually this may mean not only configuring unique virtualized sessions, but hosting them on the processor best suited for the work. Currently, Intel, AMD and even TI have announced support for this platform and it was initially announced around ARM, suggesting a level of processor diversity unknown and thought to be impossible in this market.
So this is like a mainframe given the nature of the likely customer (it will be a highly shared resource), but the massive flexibility of this system makes it as far from a mainframe as you are likely to get both inside and outside. It is a rethinking of the server, embracing both old and new requirements and packaging them into something very new and incredibly different.
This is one of the truly revolutionary moves in the server space, but revolutionary moves tend to come with a unique set of problems and requirements. This is largely why we don’t see them very often; it isn’t that vendors can’t do revolutionary things, it’s that moves like this have a cost and risk most are unwilling to incur.
In this case, HP has to now convince the market that this new approach will provide benefits that will justify the risk of doing something new and different and typically this is a very high bar. However, because much of the market that will want to use this technology is itself relatively new and more technically aggressive (many build their own custom servers -- granted this creates a unique not-invented-here problem that will need to be overcome), the bar isn’t as high as it might otherwise be.
Cloud service companies are aggressively moving to new technologies in order to compete in what has become an extremely aggressive market and should be more willing to accept something new and different than a more traditional vendor might. Still, HP will have to control the image of these new products to assure the energy, cost and flexibility benefits overshadow the typical teething problems associated with a new system.
This means the marketing requirement will at least be a magnitude greater than your typical new server line release and that HP will need third-party showcases that assure the benefit promises are worth any reported problems that come up during the initial years of system use. This is a high requirement for a company not currently known to be able to market at this level. Should this initiative fail, this is likely where that failure will be sourced.
If HP pulls this off, it will be at the center of a firestorm because this massive system architecture change will likely create a whirlpool of changes to networking systems, storage systems and management software that would be unprecedented in current times. Basically, it would be a redefinition of the data center around a new extremely flexible core architecture and that flexibility would need to be mirrored in all of the systems that support it eventually, so that they become optimized for these new systems as much as they have, over time, been optimized for the old ones. Even cooling and power delivery/backup systems may need to eventually be rethought because these new systems have lower requirements than the old, and additional savings will probably come from approaching the related problems very differently.