Arik Hesseldahl

Recent Posts by Arik Hesseldahl

In Server Sales Push, Dell Comes Out Swinging Against HP’s Moonshot

rockem_sockem_380Computing giant Hewlett-Packard has placed a lot of hope for its efforts at a turnaround in sales of servers on a new product called Project Moonshot.

But it’s turning out that Moonshot, because it’s pretty different from conventional servers, has opened HP up to some trash-talking by rivals, namely Dell. Soon to be going private, Dell has made it clear that it intends to play aggressively in every market in which it participates, including that for servers. And now, a post from an internal Dell blog gives an idea as to how aggressive it is going to be.

First, a reminder of what Moonshot is: Essentially, it’s a very small server, designed to consume very little energy. A routine HP claim is that Moonshot consumes 89 percent less energy to operate, and takes up 94 percent less space than a conventional server.

In the Dell post, shared with AllThingsD and which you can read in full below, Dell argues that HP’s claims about Moonshot are misleading. Using an HP white paper in which the original claims were made, Dell sought to make some comparisons.

As you might imagine, Dell found that instead of buying HP’s Moonshot servers using Intel Atom chips, you could do better buying a Dell server running Intel Xeon chips. Here’s Dell’s money quote: “The correct conclusion, once we strip away HP’s claims, is that Moonshot is not an improvement upon any of the Xeon metrics in currently available servers.” Basically, the new thing isn’t any better than the old thing, so buyer beware.

It’s worth noting there that the Atom chip used is a smaller low-power processor that is a cousin of the chip Intel has aimed at mobile phones and tablets, and that Xeon is its mainstream server chip.

I took all of this to HP, which declined to comment.

Update: Actually HP is hitting back. I just got this statement from an HP spokesman: “What do you expect from our competitors who do not have something as game changing or industry leading as Moonshot? Moonshot threatens the server status quo and legacy server providers, revolutionizing the economics of the data center by delivering 80 percent reductions in energy use, in 80 percent less space for 77 percent less cost.”

On top of that, HP questioned the methodology of Dell’s comparison and the results. It also points out — not unreasonably — that HP’s white paper benchmark was done using 1 gigabit network connections, which are the most common used in data centers. Dell said that in its test it used 10GB connections — which are a lot less common in rack server environments.

I also ran Dell’s claims by a few third-party analysts who deal in this sort of thing, and, while none of them wanted to go on the record on all this, they did share some observations, which I’ll summarize.

They defended HP a bit, saying that HP’s original claims were made on a first-generation system intended to test the Moonshot concept. That machine used an earlier generation of Intel Atom chip, code-named Centerton, which has since been supplanted by a newer one, code-named Avoton, that has better performance and power-consumption characteristics.

And better Moonshot boxes are coming, running not only Avoton, but also Intel’s mainstream server chip Xeon. Nor does it consider — and indeed couldn’t, because they’re not shipping yet — forthcoming iterations of Moonshot running chips based on the ARM architecture from companies like Calxeda, AppliedMicro and others. The Moonshot story, they said, will get better with time.

Secondly, the comparison only focuses on one or two kinds of workloads, essentially serving static Web pages, whereas there are many more things that servers like this can be used for, so there’s many more comparisons that can be made.

Still, this is a pretty good indicator of Dell’s in-the-trenches plans to market against HP and Moonshot in the coming weeks and months. If you’re in the market for servers and thinking about Moonshot, you can expect that Dell will want to have a talk with you.

The original Dell blog post is below. A few caveats before you read it: First, it was posted to an internal Dell blog for its sales teams, so I can’t link back to the original; second, it’s a bit on the technical side, and dives into such things as how many watts of power the servers are consuming, and the amount of space each machine takes up in a server rack. One thing is certain: It amounts to the first shot of many.

Moonshot’s misleading claims

Dell servers have the advantage over HP Moonshot systems across the board when it comes to cost savings, space requirements and power consumption.

HP published and has advertised claims about the performance, energy efficiency and potential cost savings of its Moonshot systems. We looked at these claims and the process HP used to establish the benchmarks, did some comparisons of our own and found that:

Moonshot is 94 percent more expensive than current Dell servers
Moonshot requires 215 percent more space than current Dell servers
Dell servers require 25 percent less power than Moonshot.

In a publicly-available technical whitepaper, HP compared an HP ProLiant Moonshot server with a single two-core, Intel Atom S1260 processor with a ProLiant SL230s server utilizing two six-core Intel Xeon E5-2630L processors.

HP claimed that for static webpages, “a single HP ProLiant Moonshot Server with 1 Gigabit Ethernet network interconnect displays virtually identical performance as a traditional two-processor server.”

We replicated that comparison and observed identical performance for Atom and Xeon when using a single GbE connection on both systems, specifically that:

-Using a 10GbE NIC on the Xeon system improved benchmark performance by 7x.

-Using two 10GbE NICs on the Xeon server results in a total of 11x performance improvement

HP set up its Moonshot comparison against a two-socket Xeon server that is network constrained and severely under-utilized. Since the Xeon server’s performance is limited by the gigabit network (only a single GbE NIC is used on the SL230s server), the Atom and Xeon servers show identical performance.

In the same test using a two-socket Xeon server that is not network constrained, the Xeon server performance is better by 11x. We validated this correlation in the lab, and if we redo HP’s gamed math, we find that to achieve cost and power savings over Moonshot, a customer only needs:

Two S Xeon servers for $8,000, using 5U of space and 225W of power, versus 45 Moonshot Atom servers for $62,000, using 4U of space and 720W of power.

Using this configuration, to achieve the same performance as a Moonshot server, a customer buying Xeon servers from HP will need: 45 SL230 servers, for $360,000 which require 22.5U of space and 10KW of power.

The correct conclusion, once we strip away HP’s claims, is that Moonshot is not an improvement upon any of the Xeon metrics in currently available servers. Be confident as you help customers cut through the hype that HP is putting into the market with its Moonshot claims.


Latest Video

View all videos »

Search »

There’s a lot of attention and PR around Marissa, but their product lineup just kind of blows.

— Om Malik on Bloomberg TV, talking about Yahoo, the September issue of Vogue Magazine, and our overdependence on Google