CXO

HP rethinks the server

Patrick Gray takes a look at HP's reinterpretation of the server: its new Moonshot series.

Just a few days after my blog about Data Center trends mentioned "rethinking the rack," HP announced its reinterpretation of the server with its new Moonshot series. While the name initially struck me as odd considering HP's ongoing troubles, the product appears to be a refreshing take on the problem of cramming more processing power into less real estate, with lower power and cooling requirements.

Moore power

For years, data centers benefited from the largess of Moore's law, with each upgrade demanding the latest enterprise-grade processors, and the explosion of multi-core chips and new fabrication technologies packing unfathomable levels of processing horsepower into each new generation of processors. However, powering and cooling these monsters has combined with the fact that many applications simply don't need the raw computing muscle of the latest multi-core behemoth. Power consumption has moved from a trivial number on the spec sheet to a primary consideration when designing a data center. With Moonshot, HP has simply productized a notion that's been circulating in some circles for a few years, that lower-power mobile processors are "good enough" for most enterprise computing tasks, and the ability to cram several of them into a small space outweighs any performance penalty.

Changing chassis

In some ways, the Moonshot chassis is more similar to current server chassis than something like the Open Compute Project. The OCP server eschews traditional niceties like a cover and bezel, and looks like a server with an open case. Moonshot looks rather traditional, but essentially puts an entire rack's worth of servers in a single box. The first Moonshot chassis contains shared networking, power, cooling, and storage, and allows for 45 pluggable "servers" sporting one of several available chipsets. Essentially, HP has taken the blade server to the next level.

Will HP shoot the moon?

Conceptually, HP has made a logical move, further condensing the blade concept and allowing for a "rack in a box." Employing processors with a mobile heritage allows for lower power and cooling requirements, and allows the end-user to configure a variety of ARM, Intel, and AMD processors in a single, standardized box, choosing the right tool for the job without requiring a different hardware platform. All of this sounds compelling, save for two chinks in the armor: going first and putting all your eggs in one basket.

While few components of enterprise servers are interchangeable among different vendors, the designs and maintenance procedures are relatively proven. With Moonshot, HP is offering a fairly dramatic departure from traditional server designs. If this new design is embraced, HP has the chance to define a new server architecture and reap the rewards of being the first to market. However, if Moonshot gets a lukewarm reception or is plagued by technical bugs, other heavy hitters in the server market can wait on the sidelines and release a superior product.

The other major risk to Moonshot is that it puts what amounts to an entire rack of servers into a single box, with shared networking, power, cooling, and storage. While high rack densities are nothing new, the risk of a single hardware component failure taking out 45 servers may be a point of concern for critical applications.

The market will ultimately decide whether Moonshot is the next big innovation in data center design or a niche proprietary player. In either case, it's refreshing to see a major player fundamentally reconsidering server design, rather than just throwing out increasingly power- and cooling-hungry designs.

For more on the 21st century data center, see ZDNet’s special feature page, or download TechRepublic’s Executive Guide to the 21st Century Data Center.

About

Patrick Gray works for a global Fortune 500 consulting and IT services company and is the author of Breakthrough IT: Supercharging Organizational Value through Technology as well as the companion e-book The Breakthrough CIO's Companion. He has spent ...

7 comments
sharkeys.machines
sharkeys.machines

I don't see HP putting all of there eggs into one basket. This product is not designed to replace the standard server - it is aimed at a very specific market. Nor does it have any more Points of failure than current products in the market. I do see this as being the Beginning of long awaited changes and a Company will to be strong enough to Lead.

pipervt89
pipervt89

While there are definite advantages to many, many processors handling the things they can do, enterprise licensing which is charged by core or by chip will offset some of the potential cost savings from an infrastructure standpoint. We all like lower utility costs, not only the electrons destined for the computer but the electron going to HVAC as well, but more less powerful chips means more software fees. Brian HP ASP

Systems Guy
Systems Guy

Probably not, just like existing servers, I'm sure Moonshot servers will have redundancy built in. If not, yeah, problem waiting to happen. Further, I can't imagine an enterprise installing this type of product and it NOT have at least power supply redundency, etc.

zbrkic
zbrkic

Now, the concept reminds me on Intel MultiFlex that lives for almost a decade and now is near the extinction (not for technical reasons)...

frylock
frylock

Qualitatively at least, I like the idea of higher-density/lower-power boxes. To be sure there are some challenges, like the article mentions (reliability). I think software is going to have to improve as well if it's going to run on a larger number of slower cores. Writing highly scalable mutli-threaded code is not trivial.

ChrisTheta
ChrisTheta

I agree. With Microsoft SQL Server's new licensing strategy, you want to have few of the fastest CPUs there are, because you not only license per core but there are minumum numbers of licenses you need to buy. So that $4,000 server may need $50,000 in SQL licenses minimum, even if you don't need the processing power of having all cores doing SQL.

nick
nick

Chris, you make a valid point about SQL Server licensing and a number of other applications are in the similar position. However you need to consider the whole infrastructure surrounding your SQL Server. It is becoming increasingly common to have a multi tiered approach with the SQL Server forming the back end and an array of servers in the front end and even now an array of servers in the middle. Typically the front end are web servers which have lower licensing costs which can be practically zero if you use Linux, Apache as a front end. So using Moonshot you would still have the SQL on a grunty server, as you point out, but the front end can certainly be cheaper, less powerful, more environmentally "green" servers.