HP has been in the headlines a lot over the last few months while the company tried to decide if it wanted to be in the consumer PC business. It does. But one thing that’s never been in doubt is that HP wants to ramp up its enterprise software and hardware offerings — and today the company rolled out a new enterprise program called Project Moonshot.
Basically, Moonshot is built around new low-power server technology. At launch, HP is using technology developed by Calxeda to offer servers built around low power, ARM-based processors. But HP isn’t married to ARM so much as low-power chips. In the future, we could see servers from HP using other processors such as Intel Atom chips.
The idea is that by using low power chips, you don’t just reduce the electric bill — but you can also pack servers in tighter by sharing resources. It’s easier to cool low power chips, for instance, so you can use the same cooling system for multiple servers.
HP says that allows the company to put 288 servers in a chassis designed for just four. See that picture at the top of this article? All of those little white square signify servers. Each one uses just 5 watts or less.
Clearly, we’re not at a point where all server needs can be met by this type of system, but HP is pitching the new project for companies that offer web services, social media, or other relatively simple content.
Low power chips aren’t just for smartphones and appliances anymore.
Granted this will not filter down to mid sized corporations for quite some time but for large enterprise, particularly those serving streaming media, cloud services and even data mining and warehousing will benefit from this greatly. Â These 4U boxes can be setup in any type of config from grid, mesh to individual node computing. Â Hell a standard rack would house 11,520 cores. Â So from a density standpoint it can’t be beat. Â When the Cortex A-15 goes mainstream and in particular its’ successor in 64 Bit ARMv8 is when we will see this technology really take a foothold in data centers worldwide.
Code name “Project Moonshot”? Â Reminds me of a photo I saw recently of Steve Jobs showing an iPhone, with a caption that read “Remember when ‘cutting edge’ meant moon rockets, not just a better way to post tweets?” or something like that.
I think using ARM on the server will make a lot more sense once 64-bit ARM becomes a reality. Once that happens, we can have 288 servers in a single chassis — and each server can have 64 gigs of memory, a 1tb one-chip SSD drive — and can run a dozen Xen/KVM/Vmware/VirtualBox servers, or 50-100 OpenVZ/LXC/FreeBSD Jail/Solaris Zones. So, we will soon be able to have up to 28,800 tiny virtual containers in a single chassis.Â
Of course, we will need IPv6 to make it cost-effective to give each of those containers their own IP.
I wonder why there are so few cores per chip. Oracle (fka Sun) SPARC T3 had 16 cores/chip (with 8 threads/core = 128 threads) and T4 has 8 cores/chip (also 8 threads/core = 64 threads/chip). Yes, I know, SPARC isn’t low power, but it seems likely that you’d need multiple scores (or maybe even over a hundred) of ARM Cortex-A9 cores to equal a single T3 or T4 chip on multithreaded performance.
Most don’t realize that ECC RAM is essential for servers in a production environment. Â ARM has been is servers for a long time, but this makes its presence legitimate. Â Sadly, as ARM comes “up” in applications (to stationary devices) and x86 comes “down” (to mobile ones), I don’t think ARM will end up in very good shape. Â The business model is awful. Â In fact, I still believe that MIPS may outclass ARM. Â China is pushing MIPS “in house”. Â That’s a huge market.
ARM is beginning to evolve for the server market. Soon we’ll be seeing ARM systems designed specifically for server applications that will be very different from what we’ve seen and will continue to see in the mobile market.
Though we’re still years away from seeing 64bit capabilities from ARM, even though they just introduced ARMv8 for 64bit architecture.Â