This morning, HP took a bold step forward with formally unveiling their "Project Moonshot" -- a new class of servers targeted at (apparently) hyperscale web server farms.
While I'm certainly a big fan of disruptive innovation (even from EMC competitors), I've come away with mixed emotions.
HP has certainly picked an interesting target to aim at. And they have enough resources and influence in the larger ecosystem to drive a specific agenda.
But I think it's going to take an awfully long time before the jury is in as to whether they've been successful or not.
This morning, HP conducted a press event announcing their new "Project Moonshot" servers: a new design package with new processor choices.
Strip away the fanfare and puzzling branding (e.g. "software-defined servers"?), and we see a big bet on several interesting industry trends:
- server architectures designed for web-scale processing will be an important market (think Facebook and their ilk)
- power, cooling and floor space (opex) become dominant concerns with this audience
- newer, low-power SoC (server on a chip) processors can do a better job on powering web services per $$ and per watt
Download the data sheet, and you'll see the currently ability to cram 45 Intel Atom 1260-based servers in a 4.3U rack unit.
So far, so good.
A good deal of the online commentary had to do with the mention (but current unavailability) of more radical processor designs, such as ARM.
The notion is simple: if it's all about hyperscale web farms where $$ per physical server and watt per physical server really matters -- isn't that where the real action is?
The reason is perhaps complex, but one harsh reality stands out -- the software ecosystem is years away from critical mass to make these non-x86 compatible architectures appealing for broader audiences.
Yes, there's plenty of good open source theoretically available, but unless you've got a very large in-house engineering team, the requirement to port over, qualify and support the hundreds (or perhaps thousands) of software components required is an extremely daunting task.
You're going to need much more than just a Linux distro.
The web-scale crowd may also be deterred by the limited I/O capabilities of the current design. Yes, there's room for a single 2.5" drive (either disk or SSD), and a dual-port 1Gb controller from Broadcom -- but the lack of PCIe precludes the fastest (and most cost-effective) forms of flash.
Not all that appealing to anyone with IOPS-intensive workloads.
HP's bet is also bold in that it squarely targets an emerging market; and -- as a result -- has little (if anything) for bread-and-butter IT shops where most of their revenue comes from today.
There's bound to be serious skepticism as to whether familiar enterprise-class workloads can be run on Intel Atom processors with 8GB of available RAM, as well there should be. And most enterprise shops are very comfortable (and heavily invested in) the notion of using server virtualization to make several smaller virtual servers out of bigger ones.
One should keep in mind that HP doesn't appear to be heavily invested in any server virtualization technology -- choosing instead to play the field -- are they attempting to do in hardware what's already been done in software?
The Big Bet
I do applaud HP for making a bold bet -- they're always good for the industry.
The timing of the announcement appears to be partially driven by HP's need to start to reclaim some of its mojo in the marketplace. But I don't think we're looking at the real product today -- just a predecessor of the next version.
If the next version includes more serious low-power CPUs (and some mechanism for either bus-attached or motherboard-based flash technologies), we're on to a more serious discussion: will hyper-scale webfarm operators invest the required engineering effort to bring their wonderfully complex software stacks over to a new CPU architecture?
But only the biggest outfits have the resources to play that game.