Hewlett-Packard may had been wrestling with a lot of considerations as CEOs and their suggestions come and go, however the company's server professionals understand a doubtlessly explosive company probability when they see it.
it's why HP has put together a new hyperscale company unit inside of its commercial enterprise Server, Storage, and Networking behemoth and is getting out the door first with a server cluster that makes use of the EnergyCore ARM RISC server chip simply introduced by using upstart Calxeda.
The hyperscale server effort is called task Moonshot, and the first server platform to be created below the undertaking is known as Redstone, after the floor-to-floor missile created for the U.S. army, which changed into used to launch america's first satellite in 1958 and Alan Shepard, the country's first astronaut, in 1961.
Moonshot, of path, refers to NASA's Apollo task to position a man on the moon with the aid of the end of the Nineteen Sixties, and there'll no doubt be lots of people calling this HP's undertaking Moneyshot, referring to something simply somewhat diverse. but HP's aim with venture Moonshot is larger than just making some cash peddling alternate options to x86 chips for cranky net operators who desire do to more work for much less funds and with much less juice. So don't get the wrong idea that Moonshot is well-nigh ARM.
"here is an extension of, not a alternative of, their ProLiant and Integrity server strains," Paul Santeler, established supervisor of the hyperscale business unit inside the business standard Servers and application division at HP.
And whereas HP was now not in a mood to talk specifics, Santeler stated that Moonshot would consist of tremendous-dense servers in response to Intel low-energy Xeon and Atom chips and superior Micro instruments low-vigour x86 processors as well as assorted suppliers of ARM-primarily based server chips.
HP isn't attracted to locking itself into one ARM company any further than it may possibly come up with the money for to rely on simply Intel or AMD by myself. in this world, with technical, economic, and natural failures, you need to as a minimum twin-source key add-ons, and equally greater importantly, different ARM server chip variations are going to be decent at different things.
As we've suggested elsewhere, the EnergyCore chips are one in all a handful of ARM processors that have been expressly designed for hyperscale server workloads. The ECX-1000 processors which are the first in the EnergyCore items are 32-bit chips on the way to come in two-core and 4-core variants.
They include memory, I/O, and storage controllers and an embedded Layer 2 change cloth on the chip, which means you can simply wire a 4GB DDR3 reminiscence stick with one, slap on some I/O ports, plug these toddlers right into a passive backplane, and you have got interconnected server nodes that take the vicinity of rack servers and desirable-of-rack switches.
The EnergyCore textile switch embedded on each and every chip can implement a 2d torus, mesh, fat tree, and butterfly tree topologies and scale throughout 4,096 sockets (each socket is a server node, considering that Calxeda is not doing cache coherency throughout sockets).
The original Calxeda reference design from closing yr was a 2U rack-hooked up chassis that stuffed 120 processors (and therefore server nodes) into that metal field. With the production-grade ECX-1000 processors. Calxeda has put together a 4-node server card with 4 reminiscence slots and PCI-hyperlink interconnects so it might snuggle right into a passive backplane to get energy and check with its ARM peers and make a community.
A 4-server Calxeda EnergyCard equipment board (click on to magnify)
To make the Redstone, HP took a half-width, single-top ProLiant tray server and ripped out virtually every thing but the tray. In goes the passive backplane that the Calxeda EnergyCard, and HP can cram three rows of those ARM boards, with six per row, for a total of seventy two server nodes, in a half-width 2U slot, like this:
An HP Redstone server tray filled with Calxeda ARM servers
The trays slide into the 4U version of the ProLiant SL6500 chassis, and you can put 4 of those trays in the chassis thus:
HP Redstone SL6500 chassis fully armed and bad to 32-bit parallel workloads
That offers you 288 server nodes in a 4U rack space, or seventy two servers per rack unit. it truly is 20 per cent greater server density than the alpha check machine from Calxeda might do past this year with very early samples of its ARM chips.
That SL6500 chassis within the Redstone gadget has three pooled vigor materials that may lower back every other up and hold the nodes getting in the even one of them goes the manner of all flesh. The equipment has eight cooling enthusiasts. every tray has four 10Gb/sec hyperlinks that come off the interior EnergyCore material switch.
All of those ports may also be cross-related the usage of 10Gb/sec XAUI cables, and scaled throughout as many as 4,096 sockets. (incidentally, four,000 servers is fairly a whole lot the upper scalability limit of a Hadoop cluster these days.)
despite the fact, the advised configuration at first should be to use hyperlink the seventy two 4-server nodes in a single SL6500 to each other with the built-in textile swap, and then glue dissimilar SL6500s to each other using a pair of 10GE appropriate-of-raw switches.
In effect, the SL6500 is the brand new rack, with an built-in right-of-rack swap, and both exterior 10GE switches are equivalent to an end-of-row change that continually hyperlinks numerous racks to every different.
That 288 server count on the Redstone gadget assumes that you're going to network out to exterior disk arrays, but you could sacrifice some servers in the trays and plug in up to 192 solid state disks or ninety six 2.5-inch disk drives into an enclosure. The SSD and disk pressure cartridges plug into SATA ports on the EnergyCard and draw their energy from the backplane in the tray.
The earnings pitch for the Redstone techniques, says Santeler, is that a half rack of Redstone machines and their external switches implementing 1,600 server nodes has forty one cables, burns 9.9 kilowatts, and costs $1.2m.
A greater usual x86-based mostly cluster doing the identical quantity of work would best require 400 two-socket Xeon servers, however it would take in 10 racks of area, have 1,600 cables, burn ninety one kilowatts, and value $three.3m. The big, massive caveat is, of direction, that you want a workload that can scale neatly on a modestly clocked (1.1GHz or 1.4GHz), four-core server chip that most effective thinks in 32-bits and handiest has 4GB of memory.
"There are a lot of valued clientele that I even have talked to who feel 32-bit is simply best," says Santeler. The chips will likely be decent at net serving, net caching, and big information chewing workloads the place processing records in smaller bits is the norm, no longer the exception.
That referred to, HP does not seem to be in a big hurry to commercialize the Redstone machines, but it is getting machines accessible as Calxeda is starting to do samples – and that is about as good because it can also be.
The ECX-one thousand chips from Calxeda are anticipated to pattern late this yr, with volume shipments within the core of the yr. HP's Redstone machines the usage of these chips could be obtainable in constrained quantities for a limited number of customers within the first half of 2012.
And here's the kicker: Santeler is asserting that HP is making no commitments at present about when it is going to ship as a often accessible product, or even if it's going to. (I feel HP is only being overcautious and dramatic.)
HP is placing Redstone machines into DiscoveryLabs around the globe, beginning in a data middle in Houston, Texas, the place its computer and server manufacturing unit is, and may provide abilities valued clientele a chance to add their classes on the Redstone servers and put them through the 32-bit ARM paces. And for now, that means doing so on Canonical Ubuntu or purple Hat Fedora Linux, which support ARM chips and which had been tweaked to assist the Calxeda chips.
incidentally, HP has no plans at the moment for placing ARM processors in its commonplace goal ProLiant DL rack, ProLiant BL blade, or ProLiant ML tower servers. ®
subsidized: fitting a practical safety chief