There are a number of ways to create a power-efficient server chip for hyperscale applications like those that run at Google, eBay, Facebook, and so on. One is to rip damned near all of the guts out of a Xeon processor and make an Atom chip, which Intel has done. Another is to beef up a MIPS or ARM RISC processor aimed at …
RISC servers = lovely!
With our "No source, no sale" procurement policy, we'll be well placed to take advantage of cheap, economical-to-run RISC based servers.
Best of luck to them
Best of luck to them. If you've got some software that runs on Linux, why should it have to be Linux/x86 simply because it needs to fit in a "server" form factor ?
"Because that's what the IT Department's comfort zone (Window boxes) has been for the last ten years" is a common (albeit appalling) answer. It's time to change it, and save on electricity, and cooling, and and and.
our new sling shot wielding small overlord.
I wish them success. They need marketing and political luck as much as a good design.
good for cutting cost of professionally hosted servers
I'm not aware of any software on the current shared X86 servers split into virtual machines which I administrate which doesn't compile to run on ARM. This kind of thing could put a dedicated server within my small hosting budget. The applications are more likely to be RAM bound than CPU bound. Also a large part of the machine rental relates to the power requirement. So bring it on, as far as I am concerned.
Bring it on!
A non-x86 arch is what we should be using, and that should've happened years, if not a decade ago. x86, even in 64-bit mode, is a steaming pile of crap. Intel basically won because Windows won the OS wars, and Windows ran on x86. Had RISC OS or MacOS (Classic) won back then, we would all be running RISC machines, maybe even true 64-bit by now.
I wish these guys good luck .. maybe, just maybe they might take on two goliaths at the same time: Intel *and* Microsoft. Go David Go!
This'll be interesting.
ARM is kind of a ninja architecture. It's quiet, efficient, and all over the place, yet most people are entirely unaware of the fact--but Intel had better glance over their shoulder now and then or it might sneak up on them while they're out for a walk. Intel's used to going up against chip companies with lower production volumes and higher production costs than they have--but is that really what they're up against now?
This one server startup probably looks small, weak, and financially inconsequential next to the x86 server business, but that might be a dangerous misjudgement.
Get in there
So little server work is high-CPU, it's about time ARM got a little more of the action.
I'll take issue with this though - "Alternative power-efficient Power processors from PA Semi disappeared into the gaping maw of Apple, never to be heard from again."
The demise of Apple's use of PPC chips was down to no-one with PPC design facilities being interested enough to create low power G5-class chips, leaving Apple lagging badly behind in laptop speed, stuck with G4s, from about 2004.
PA Semi did get into that field, but the initial PWRficient (gak!) samples in 2007 were a year or two too late to meet Apple's timelines - fully a year after Apple announced the switch to Intel - but the chips went out to other customers instead.
TI quit making chips entirely
"Other Smooth-Stone backers include chip maker Texas Instruments (which used to be the foundry for Sun Microsystems, but lost that job to Taiwan Semiconductor Manufacturing Corp) ..."
TI did not "lose" the job of making SPARC chips to TSMC, they quit. TI exited the semi-conductor fab business, as they judged that it was too expensive to go past the 65nm process. In fact, I believe The Register wrote about it.
mips per watt's where it's at
I've installed *desktop* linux (ubuntu, debian, or gentoo) on alpha, sparc, hp pa-risc, powerpc, x86 and x86-64, and other than physically looking at the box (or looking at /proc/cpuinfo or dmesg), there was no way to tell which platform i was on. Well, except no flash (they did have gnash though). Run the package manager and there's a *full* set of packages. Server? No sweat... My point being, you won't find you're missing packages you need for your server because you leave x86 behind... If you can get the job done and save power by leaving x86, there's no reason not to.
That would solve a lot of issues. Lots of other interesting lines to pursue here.
Intel's been working on getting the watts down though with Atom. Maybe they've got a few interesting options along this line. Competition is a good thing.
It seems the MIPS architecture is going back to its roots... It always was a server platform, and at one time was the king of floating point performance.
I personally can't wait for ARM based servers, everything i'm running is open source, is already compilable for ARM and the majority of my hosting cost is the price of power these days.
Yes, of course "Linux" runs on ARMs of all kinds, but for a credible alternative to x86/x64 servers, the IT departments will be looking for an industrial-strength distro like RHEL or SLES or maybe OEL or CentOS. I wonder if Smooth-Stone (not to be confused with Smoothstone, an entirely different company - www.smoothstone.com) are talking to the big distro players?
RHEL / SLES
Actually, people only choose RHEL / SLES because those distros support binary blobs. (CentOS is *just* RHEL, recompiled with different default graphics in place of Red Hat's trademarks.)
No (x86) binary-only anything will work on ARM, so with any luck we can start with fully i-tal drivers from day one and stay that way. I expect Gentoo (which is source-based, and so can be installed on anything with a port of GCC) and Debian (which adheres to Free Software ideals like a neodymium magnet) to become the dominant distros on ARM.
What you really need...
What you really need is for the commercial software houses to create software for non-x86 architecture. It's all very well if you can run the FOSS software but most companies rely upon some commercial software - backup clients would be a classic.
If the commercial software houses would simply *release their Source Code*, then they could find their software installed on non-x86 architectures.
But that's ultimately a decision only they can make (barring legislation requiring them to do it).
PS. Bonus question: Why does anyone need anything more than cron, find, tar, bzip2 and scp for making backups anyway?
Assuming that we're always going to be stuck with closed source commercial software and sorry to keep banging on about Storage but that's mainly what I do at the moment - I've yet to see a FOSS backup system which is worth more than a cursory look. The new platform will require commercial companies to compile their software for ARM.
I am an ARM fan and I am a linux fan, I'm just being realistic here - the killer commercial app (whatever it is for whichever company) already works on Intel, to get a foot in the door ARM servers will have to have whichever app that it - it varies for each company - compiled on ARM.
Because not everyone only runs one or two linux boxes that they're happy to gaffer'n'string together with homebrew scripts. Many companies run thousands of mixed OS machines across multiple sites and require central management consoles, access to all the facillities of really big storage systems, encryption, key management, media tacking, media ageing, etc. etc. etc. You just can't do this with a few scripts and cron.
The vast majority of big companies won't even use cron because it's so hard to manage on more than a couple of machines. Just think of the hassle involved if you want to stop your 10pm backups if there is a thousand machines each kicked off by CRON?
Is Symantec Backup Agent for Linux any good?
Hey Fraser, is the Symantec Backup Agent for Linux any good? The one that runs on Red Hat, SuSe, Solaris, and a handful of others already (including SPARC, so it's already relatively portable).
Does it provide any of the facilities you desire?
I know it's called Symantec which really puts me off but I used to know folks who liked it.
How long will it take for Symantec to introduce an ARM-specific version for the leading ARM Linuxes?
Just askin, like. Similar questions for other commercial products, I just cba right now.
Once one or two of the commercial players head down this road, how long before the rest are forced to follow or lose market share?