Beta
×

Welcome to the Slashdot Beta site -- learn more here. Use the link in the footer or click here to return to the Classic version of Slashdot.

Thank you!

We are sorry to see you leave - Beta is different and we value the time you took to try it out. Before you decide to go, please take a look at some value-adds for Beta and learn more about it. Thank you for reading Slashdot, and for making the site better!

Interviews: Ask Technologist Kevin Kelly About Everything

timothy posted more than 3 years ago | from the if-forrest-gump-were-smart-and-ambitious dept.

Earth 135

Kevin Kelly has for decades been involved in some of the most interesting projects I know about, and in his roles as founding editor (and now editor at large) of Wired Magazine and editor of The Whole Earth Catalog has helped spread the word about many others. Kelly is probably as close to a Rennaisance man as it's possible to be in the 21st century, having more-than-passing interest and knowledge in a range of topics from genetic sequencing and other ways that we can use measurement in pursuit of improved health to how technology is used and reused in real life. Among other projects, he's also the founder of CoolTools, which I consider to be (unsurprisingly) the closest current equivalent to the old Whole Earth Catalogs. (Disclaimer: I've had a few reviews published there, too.) (He's also one of the founders of The WELL, now part of Salon.) Kelly is also Secretary of the Board of Directors of the Long Now Foundation, the group which for years has been designing a clock to ring on 10,000 years in the future. Below, ask questions of Kelly, bearing in mind please the Slashdot interview guidelines: ask as many questions as you want, but please keep them to one per comment. He'll get back soon with his answers.

Sorry! There are no comments related to the filter you selected.

hmmm (0)

Anonymous Coward | more than 3 years ago | (#36877644)

now combine that with low power cooling that google's using, hook it up to a nearby solar power plant utilizing a salt silo as heat storage, and now we're talkin!

Re:hmmm (2)

rbrausse (1319883) | more than 3 years ago | (#36877908)

instead of the not production-ready salt silo hydroelectricity seems a more realistical solution - either as pumped-storage to balance the fluctuating production of photovoltaic/wind or run-of-the-river as 24/7 supplier. the latter is the only source of energy for one quite big hosting company here in Germany

Mozilla isn't using "the cloud"? (-1)

Anonymous Coward | more than 3 years ago | (#36877670)

How quaint.

difficultiesd (5, Funny)

bbk (33798) | more than 3 years ago | (#36877698)

the SeaMicro servers handled the load with no difficultiesd

Hmm... now there's a daemon you really don't want to see running...

Re:difficultiesd (0)

Anonymous Coward | more than 3 years ago | (#36877934)

If you don't want it, just vim /etc/default/difficultiesd and change ENABLED to 0. Then
/etc/init.d/difficulties stop
and you won't see it again.

Re:difficultiesd (0)

Anonymous Coward | more than 3 years ago | (#36878280)

Won't work. You messed up your syntax (missing 'd'). The command is:
#/etc/init.d/difficultiesd stop

You command just produces an error:
bash: /etc/init.d/difficulties: No such file or directory

Re:difficultiesd (1)

Hylandr (813770) | more than 3 years ago | (#36880558)

This is a tried and true method to use less power per server.

vim /etc/inittab

find :

id:5:initdefault:

and change to:

id:0:initdefault:

Reboot server. - Done.

- Dan.

Augh, units... (0)

Anonymous Coward | more than 3 years ago | (#36877724)

The SeaMicro production servers used just 0.02 watts of electricity per browser request, Fitzhugh said, compared to 0.17 watts per request for the HP C7000 blade servers they replaced.

Augh ... seems about what it should be (in joules).

DEC patents bloooming after Intel stole/bought. (1, Interesting)

Anonymous Coward | more than 3 years ago | (#36877750)

I was looking at my DEC Alpha today and this story came up, and I pondered on how Digital could have ever been replaced by Intel and AMD after being aquired and stifled by not one but two aquisitors in the likes of Compaq and HP. Here was 1997, and Digital already had a 64-bit 1GHz DEC/Alpha on the EV6 line of lab test and despite being an American-fabricated semiconductor from an all-American company then how was it that now the industry is destitute to ever again fabricate it's own semiconductors ever since The United States brushed udner the rug all the property theft and anti-Trust activities of primarily Intel and HP? Intel was the most un-original chip manufacture, most of it's technology bought since it was founded as an investor-startup to muscle it's way into the market like how Microsoft used XBox at a loss to enter an unrelated market.

My point being is that the heavy hitters of the industry are all prime examples of when superior American work was destroyed, and in the occasion of the DEC/Alpha it was so-far advanced that Intel slowed-down the industry using Moor's Law so they can reap the profit of every chip revision that is nothing more than the re-implementation of a crippled inferior design.

Digital was responsible for many of the ARM embedded chipsets, the greatest of Ethernet LAN chipsets, multi-processor scalability, was first to have a retail 64-bit 1GHz processor for sale in 1999 despite being in the rounds since 1997 to developing engineers, and we get stuck with foreign-made non-American monopolies. Do I need to remind everyone that Intel puts STATE OF ISRAEL on the map like what the musical-band U2 does to put Ireland on the map? Who is destroying America supremacy to push our technology into foreign countries that have never done any good to mankind? It stinks of pay-offs.

And now that The United States is in charge, America is being liquidated.

Before anyone says Intel was cheaper, look here. (0)

Anonymous Coward | more than 3 years ago | (#36877910)

http://farm4.static.flickr.com/3027/2503059877_76dc4a82e0.jpg [flickr.com]

The above image shows a Tandy 20MHz 386 computer for over $8k dollars U.S. Sure, many will say that DEC was the $10k Unix system, just consider that Intel was built on the success of a Fairchild employee Federico Faggin that created the 4004 micro-processor and this is where the ZiLOG Z80 was spun-off from while Intel went and made the 8008. So there you have it: American companies, outsourced over time, and then when these companies are in their greatest phase they outsource yet again.

Something should be done about outsourcing. Why help a bunch of thieves and murderers around the world when America can sit tight and look pretty for a couple hundred years for sake of not being on it's own island away from the world disease?

Re:DEC patents bloooming after Intel stole/bought. (1)

RightSaidFred99 (874576) | more than 3 years ago | (#36878616)

Oh, good. The Alpha fanboys have made an appearance!

What is with you old farts? The Alpha was a good chip at the time, but that was a long fucking time ago. Intel took the goodness out and it's in other, much superior designs now.

Mostly, those of us who aren't old farts are sick of fucking hearing about the god damn Alpha. Jesus Tapdancing Christ.

Re:DEC patents bloooming after Intel stole/bought. (1)

PlusFiveTroll (754249) | more than 3 years ago | (#36879358)

Actually didn't AMD get the goodness out and come up with hypertransport and a much better systems bus then Intel had for years?

Yes, HyperTransport is successor to EV6 bus. (0)

Anonymous Coward | more than 3 years ago | (#36879726)

And just to point out a little-known fact, the recent is known as the Alpha 21464 and is not discontinued but is on stand-by when Intel and AMD to provide enough bell-ringer advertisements for their recent batch of chips.

Seriously, it's verry sad that Alpha is not being manufactured except to succeed the ES and GS servers from HP, and it's the only chip on official "stand-by" mothball status. I find it odd that the marketers can actually advertise the close of a complete superior line of processors as though it's a legend only waiting for the right environmental factors to occur for it to go back to work again like God. Who? What? Why would any company continually dispense inferior technology to consumers when this verry successor to the 21364 is on wafer-waiting manufacturing-standby? I's verry design is based on a prior design based on a prior design based on a prior design -- it's a continum of the old 21164, only adjusted and re-fab'd in a tighter package, and it's wattage is actually less than the recent chips from Intel and AMD or roughly under 100-watts.

At the moment, IBM has the chore of fabbing Alpha and the prior company that was fabbing it was Samsung, but IBM is too busy with Power5 and Samsung is doing something or other.

Re:DEC patents bloooming after Intel stole/bought. (1)

Ex-MislTech (557759) | more than 3 years ago | (#36880356)

And now that The United States is in charge, America is being liquidated.

Watch the youtube video" confessions of a economic hitman", its worse than most know...

We talk about this need a lot at work. (2)

toonces33 (841696) | more than 3 years ago | (#36877762)

We have quite a few machines in the server room, and we have constant problems keeping the room cool. But ultimately many of the boxes really don't need that much CPU power - they have a fairly simple job that they need to do. We have speculated about using an old laptop on AC power for some of the jobs that don't require a lot of CPU and don't require a lot of disk space.

These servers sound like they would work quite a bit better for this purpose however..

Re:We talk about this need a lot at work. (1)

Anonymous Coward | more than 3 years ago | (#36877918)

Isn't that the use case for consolidating your servers into less hardware using virtual machines?

Re:We talk about this need a lot at work. (0)

Anonymous Coward | more than 3 years ago | (#36879122)

It can be that. It can also be the use case for spreading your virtualization server into separate machines. It's all about net workload/net power -- net workload reflects % idle, and net power does as well. Of course, work scales from 0 to max, while power scales from some nonzero idle power to max, so there's a penalty for idle CPUs -- loading all your machines to the max seems like the fix, and virtualization is how to do that. But if the machine with horsepower to run everything takes sufficiently more power at full load, you may be better off with the idle penalty of running separate, more efficient, servers.

(Of course, in an ideal world, everything is virtualized and parallelized, running on a cluster of power-efficient machines, and if your workload is predictably variable, you can just shut down a whole rack when you're not using it -- you can get almost no idle penalty this way. But we all know not everything parallelizes equally well.)

@GP: depends how much horsepower you need, but if you were considering roll-your-own on a laptop, you might look into rolling your own on an off-the-shelf single or dual-core ARM system. You can get last year's Linux handhelds, tablets, etc. for reasonably cheap these days, or a BeagleBoard for ridiculously cheap. Run Debian, be happy -- they're all significantly more efficient than any Atom machine, laptop or server. (Can't really say how they stack up against Tilera).

Re:We talk about this need a lot at work. (1)

LoztInSpace (593234) | more than 3 years ago | (#36880902)

Genuine question:
Why is virtualisation so good? We have
- Multi-tasking o/s that generally have resource allocation/partitioning/caps etc
- Database software that runs multiple instances/databases/schemas
- Web servers that run multiple sites on multiple threads
So why is virtualisation better than stuffing your physical servers to capacity rather than adding overhead of multiple o/s
Seems like the logical end-point will be a single-process O/S run off a mega hypervisor, which is almost indistinguishable from an O/S

Re:We talk about this need a lot at work. (1)

Arrepiadd (688829) | more than 3 years ago | (#36880932)

The reasons for this vary dependent on who is using them and why, but one possibility is someone wanting the mail server on a separate machine than the web server, yet working on a company small enough that it doesn't make sense to have a physical machine for each (in the sense of too much money/power consumption for a machine that will hardly do anything)... You virtualize the two machines and put them running on an actual server.
This allows you to have different access levels and security restrictions for each of the machines, while keeping the costs minimal (just one machine and the electricity to keep that running).

Re:We talk about this need a lot at work. (1)

LoztInSpace (593234) | more than 3 years ago | (#36881168)

Thanks. While I agree it's a possible or even likely scenario, I'm pretty sure virtulisation in this case would be an unnecessarily wasteful solution. Don't most O/S s allow you to apply security at the application/service level? I'm not saying you're wrong in that people think like that, but I'd expect admins to know that it's not the only way.
Anyway, thanks for the reply.

Re:We talk about this need a lot at work. (1)

asdf7890 (1518587) | more than 3 years ago | (#36881418)

You could manage security at the per service level, but often managing it at a per server level is slightly less hassle, so the VM option reduces admin effort a little in that area.

Also by keeping the tasks on separate (virtual) machines you reduce the chance of a configuration change for one app having unexpected effects on another. Shared libraries can be an issue to - some commercial app vendors will not support you if you don't have exactly the right library versions and those exact versions might not be the same as the versions required by another app. While it is possible to have multiple versions around on one OS instance, using VMs is much less hassle.

There are other benefits to: for instance when the workload seen by one or more of the tasks increases to the point where the host is overloaded, moving a VM to a new host is often quicker and easier than pulling a particular service from one mixed task machine to another.

You are not wrong in that the overhead of virtualisation can be considerable and in some cases the option is far from optimal, but for a lot of environments it ends up making sense to use it over other options.

Re:We talk about this need a lot at work. (1)

me.at.work (249034) | more than 3 years ago | (#36881484)

For us the reasons to go virtual (since 2006 iirc), in no particular order
- reduce power usage and cooling
- less hardware to manage
- more efficient use of hardware
- faster deployment of new servers
- separation of applications (mainly a problem on Windows, applications simply don't co-exist in an nice way on the same server)

We're not a major player by any standards but we now have 10 hosts running a total of 380 vm:s. Lot's of room for more vm:s in there still but it's designed for redundancy, 5 of the hosts can fail and all vm:s can still be kept running on the remaining hosts.

Re:We talk about this need a lot at work. (1)

vegiVamp (518171) | more than 3 years ago | (#36882168)

Host clustering, for one - you make a cluster of host servers, and you can move the guests between them as necessary, in most case with not a single packet lost - yes, that's live migration to another physical machine.

You can even keep a live copy of a running VM in sync on a different physical host, so that even if a physical host crashes, the service can transparantly flip to the copy with everything intact, including in-flight transactions. Pretty impressive, really.

And, of course, capacity expansion is transparant - you add extra hosts to the cluster and redistribute the guests on-the-fly.

Re:We talk about this need a lot at work. (1)

drsmithy (35869) | more than 3 years ago | (#36882380)

Why is virtualisation so good?

Fundamentally, because it's easier to manage a bunch of machines doing one thing each, than one machine doing a bunch of things.

Re:We talk about this need a lot at work. (1)

jmitchel!jmitchel.co (254506) | more than 3 years ago | (#36883578)

It is simply easier and less fiddly to manage applications when they are all totally isolated from each other. I've found that it's pretty damn complicated (especially if your dumb ass sales guys haven't negotiated a proper maintenance window) to negotiate downtime across multiple customers or constituencies. Applications step on each other, have subtly different requirements and expectations, and generally expect to be the only thing running on a system. For instance you stick multiple applications in that one database - and then one application requires you to upgrade and one requires you to not upgrade. And the same goes for web servers - suddenly you find yourself stuck on Apache 2.0 because you require some crufty plug in.

All of this can be worked around with a bit of cleverness and perseverance but hardware is cheap and good IT guys aren't.

Re:We talk about this need a lot at work. (0)

Anonymous Coward | more than 3 years ago | (#36878160)

Why an old laptop?

If it is critical get something high quality.

Re:We talk about this need a lot at work. (1)

Cramer (69040) | more than 3 years ago | (#36878662)

Because you can often do in a laptop with 100W what you cannot do in a 1U server with 300W. A laptop is a much more efficient design -- it's the next best thing to a blade server.

Re:We talk about this need a lot at work. (1)

martinX (672498) | more than 3 years ago | (#36880068)

http://www.apple.com/au/macmini/server/specs.html [apple.com]

85W. And you can get a rack for them. http://www.sonnettech.com/product/rackmacmini.html [sonnettech.com]

Re:We talk about this need a lot at work. (2)

dave420 (699308) | more than 3 years ago | (#36880126)

They are hardly servers. Not even close.

Re:We talk about this need a lot at work. (1)

Antarius (542615) | more than 3 years ago | (#36880436)

They're less of a server than a laptop is?

Re:We talk about this need a lot at work. (1)

dave420 (699308) | more than 3 years ago | (#36880458)

An old laptop? Of course. I don't know why you'd want all the media-based nonsense in a server.

Re:We talk about this need a lot at work. (1)

Dog-Cow (21281) | more than 3 years ago | (#36881582)

What media-based nonsense? You sound like a bigot. The mini has the same power as any laptop, but in a better form factor for non-mobile use.

Re:We talk about this need a lot at work. (0)

Anonymous Coward | more than 3 years ago | (#36887968)

We switched to mac minis since idiots kept closing the lid of our firewall.

Re:We talk about this need a lot at work. (1)

cgenman (325138) | more than 3 years ago | (#36878402)

Like RAM in the 90's, the limiting reagent for a lot of server rooms is the strength of your AC. I wonder if we'll see major leaps forward in computer cooling, or oil-bath server rooms, or server rooms perpetually doused in extra dense gas.

Re:leaps in cooling (1)

Flyin' Low (597865) | more than 3 years ago | (#36884930)

I wonder if we'll see major leaps forward in computer cooling, or oil-bath server rooms, or server rooms perpetually doused in extra dense gas.

I'm pretty sure bringing liquid cooling to the server room on large scale is the next thing in this arena. Immersion in a dielectric coolant is one way to go. I've worked with Hardcore Computer and their Liquid Blades a bit, and am optimistic: http://www.hardcorecomputer.com/servers/liquid-blade/index.html [hardcorecomputer.com]

Re:We talk about this need a lot at work. (1)

caseih (160668) | more than 3 years ago | (#36879232)

Virtualization can be an immediate, cheap solution. Since you don't need much CPU on each server, just stock one with ram and put all the servers on one box. We cut our server room from 20 or more machines down to just 8, each running 4-6 virtual machines. We backup VMs between hosts in case of hardware failure. In 4 years it has run amazingly well. Combine virtualization with low-power computers (multi-cores) would seem to be a winning combination.

Re:We talk about this need a lot at work. (0)

Anonymous Coward | more than 3 years ago | (#36880072)

You can either turn them into VMs, or just crank the clock speed down and set the disk drives to sleep on idle. If you halve the clock, the power will halve too!

512 Atoms in 10U (2)

serviscope_minor (664417) | more than 3 years ago | (#36877784)

512 atoms in 10U doesn't compare that favourable to 480 opteron cores in 10U (standard 1U, 4 socket 6100s). The atoms draw (apparently) 2.5Kw. That sounds a little low: that's about 4W each. That's plausible for just the chips themselves, but what about the RAM, etc?

By contrast, the opterons will have a 1kW PSU each for a maximum power draw of less than 10kW, which is 4x as much.

So, is a 2.3GHz opteron core 4x faster than whatever atom cores they use? Quite probably. Though they might use dual core atoms, in which case there are 1024 cores which swings it in favour of the atoms again.

Basically, the article is far too light on details.

But as always, vast arrays of weak processors is likely to be popular in some applications and be massively overhyped in others.

The atom isn't an especially efficient CPU. It's low power for x86, but the high end processors have to be very efficient to fit within the thermal envelope.

Re:512 Atoms in 10U (1)

Daniel Phillips (238627) | more than 3 years ago | (#36878128)

The atom isn't an especially efficient CPU.

Indeed. Subjectively, my dual ARM Xoom delivers far more compute power while eating less power than my Atom 450 netbook (Dell mini 10). This gives me hope for the future of the desktop, which as of today usually entails large amounts of heat and noise. In my ideal scenario, Android gets a big chunk of the tablet market, so vendors come up with the novel concept of Android desktop machines. Soon they discover that users are wiping Android and putting on standard Linux in order to run real desktop apps. So the vendors offer preloaded Ubuntu, which by this time has learned about touchscreens and how to run Android apps.

The community certainly will do this even if the vendors don't. Android may be nice for consuming media and running dumbed down games, but for nearly everything else it does not compare to a standard desktop. (Multiple windows? ... )

Re:512 Atoms in 10U (0)

obarthelemy (160321) | more than 3 years ago | (#36878396)

I actually have more faith in Android graduating to WIMP than on Ubuntu getting things right. Android seems to be run as a business, with customers needs wants and likes paramount. Ubuntu seems to be run like some hacker's project, where the latest tech shiny wins the day, regardless of whether it's actually useful, user-friendly, feature complete, and reliable. Cases in point: Grub2, Unity...

My one worry is Google limiting Android in the desktop space to leave room for Chrome OS, even though Chrome OS seems, at best, only Entreprise: no way I'm buying the HTML flavor of a dumb terminal for my personal use.

Re:512 Atoms in 10U (1)

datapharmer (1099455) | more than 3 years ago | (#36878904)

...only if you run cyanogenmod. Have you seen the crap support from venders?

Re:512 Atoms in 10U (1)

grumling (94709) | more than 3 years ago | (#36879184)

Android's biggest failure is the committee-style update process. Google is running on a ship early, ship often approach and running head long into the ultra-conservative, test, test and test some more attitude of the handset makers (who dump their crapware on) and the phone service providers (who dump their crapware on). So Google's buggy code ends up on handsets and never goes away, even though they updated and fixed the bugs months ago.

I realize the last thing the handset manufacturers want is to end up selling a commodity product, but there's still plenty of variation in hardware to keep the marketing dept happy without having to muck up the OS. The first handset manufacturer who announces stock Android and allows updates direct from Google will get my dollar. And my next phone if that doesn't happen will be a Nexus-3, as long as the hardware is better than the Nexus-S (need to have an SD slot).

Re:512 Atoms in 10U (1)

obarthelemy (160321) | more than 3 years ago | (#36881122)

I think this is more about PC vs network: Nobody cares if a handset crashes, which is what Google, and the users, deal with. Even I don't care if my telephone crashes (which happens about 1 a month). Carriers are more afraid of one buggy device, or a whole series of buggy devices, bringing down / overtaxing their network. This justifies the carriers' validation process. Not the crapware !

Re:512 Atoms in 10U (1)

grumling (94709) | more than 3 years ago | (#36885110)

Yep, I'm amazed at how many times a day the Internet crashes because of buggy devices. You'd think someone would do something about it.

That's their excuse, not their reason.

Re:512 Atoms in 10U (1)

grumling (94709) | more than 3 years ago | (#36879124)

Android has a very, very long way to go before it can be seriously considered for content creation. Not talking about applications, but just simple things like knowing when I have an external keyboard connected, don't pop up the on-screen keyboard. Cursor navigation is also just terrible, although I'll know more tomorrow when I pick up a Bluetooth mouse to see if that helps.

And just simple things like having a "home" button on a browser... Does someone have a patent on it or something?

So we throw Ubuntu or whatever on a tablet and find out that we have to recompile everything since it was written for X86 processors. Then we find out that there's a bunch of libraries that also need to be built and compiled for ARM. No way anyone but the most dedicated will even attempt it, and you'll never get those IT guys with MS Certs to even make the effort.

Seems we take a step forward on hardware and 3 back on software (but at least we got a micropayment system for software that actually works!).

Re:512 Atoms in 10U (1)

bemymonkey (1244086) | more than 3 years ago | (#36880522)

"Subjectively, my dual ARM Xoom delivers far more compute power while eating less power than my Atom 450 netbook (Dell mini 10)."

Are you sure about that? I love ARM chips as much as the next guy (heavy Android user), but I get the feeling that most of the processing (be it rendering web pages or whatever) is offloaded to the GPU and other supplementary chips... last gen's ARM chips were barely fast enough to render DivX content in software.

Are there even any tasks on tablets that are particularly CPU-intensive? There's not much in the way of multitasking (what with background apps being hibernated and so on), and single apps usually just peg the CPU until they're done... sure, iPads/iPhones and Android devices are pretty snappy, but they're still severely limited by processing power, especially in comparison to chips other than Atoms - which are also getting more and more efficient.

I bet if you put a dual core ARM in a netbook and souped it up to the same raw CPU power as an Atom, it'd draw pretty much the same amount of power and require as much cooling... Or am I missing something? Links for self-study would be much appreciated in that case ;)

Re:512 Atoms in 10U (0)

Anonymous Coward | more than 3 years ago | (#36878200)

How many dual ARM cpus could you cram into 10U?
They're just saying lower power cpus like ARM will end up hurting Intel in server farms/cloud/whatever.

Re:512 Atoms in 10U (1)

TubeSteak (669689) | more than 3 years ago | (#36878234)

But as always, vast arrays of weak processors is likely to be popular in some applications and be massively overhyped in others.

Tilera uses an architecture that eliminates the on-chip bus interconnect, a centralized intersection where information flows between processor cores or between cores and the memory and I/O. Instead, Tilera employs an on-chip mesh network to interconnect cores. Tilera says its architecture provides similar capabilties for its caching system, evenly distributing the cache system load for better scalability.

Interconnect speed has been a longstanding problem for networking.
Tilera is opening up the useful bandwidth on the CPU and surprise! they can do more work per watt.

Re:512 Atoms in 10U (0)

Anonymous Coward | more than 3 years ago | (#36878236)

Pretty much any recent ARM Cortex-A beats the crap out of Atoms on MIPS/Watt basis, and MIPS cores are not far behind.

Re:512 Atoms in 10U (0)

Anonymous Coward | more than 3 years ago | (#36878254)

The data sheet on SeaMicro's website says that they're using dual-core atoms, though it likely describes an older model since it only lists 384 physical CPUs. But I think the point of this isn't necessarily that you just want to reduce the power usage, it's that you can afford to sacrifice CPU to achieve that. Most web servers out there are ridiculously over-spec'd when it comes to CPU. RAM, network and software concerns like available sockets and file handles become the limiting factor of most well-designed web applications (note: not the db or middleware layers, just the web front-ends.) If you can spec down the CPU to save power, why not do it?

Now I'd prefer to see something like this made with ARMs (the chips...no, I'm not envisioning a server with appendages that can automatically unplug/plug-in cables and such), but atoms are a good place to start.

Re:512 Atoms in 10U (1)

wagnerrp (1305589) | more than 3 years ago | (#36878886)

Then fix the software, to allow you to take full advantage of the hardware.

Re:512 Atoms in 10U (1)

drhank1980 (1225872) | more than 3 years ago | (#36878302)

http://www.seamicro.com/node/164 [seamicro.com] Here is the Seamicro page on the system. Its using the dual core Atoms.

Re:512 Atoms in 10U (4, Informative)

wagnerrp (1305589) | more than 3 years ago | (#36878870)

They're using 256 dual core Atoms, for 512 cores at 1.6Ghz. Those 2.3Ghz Opterons will do roughly twice the work per core as the Atoms, with eight cores per chip, four chips per box, a 1U box will replace roughly 32 Atoms, requiring 8U to achieve the raw power of the 10U SeaMicro box. Those 32 processors run 80W ACP each, so including memory, disk, and chipset, you're looking at 3-4kW under load, versus 2-2.5kW for the SeaMicro.

So how much will this thing cost you? The CPUs are $500 apiece. $300 for a 1U case. $800 for a board. $60 for a 4GB stick of registered ECC DDR3, times eight per processor. $275 for four 120GB SSDs. You end up around $6K per box. Now the SeaMicro uses Infiniband for its internal networking, for communication intensive tasks. Lets do the same. $900 each for dual port 40Gbps cards, and another $6K on a 36-port QLogic switch.

That adds up to just over $61K, versus $150K for the SeaMicro box of roughly equivalent performance. For nearly three times the cost, you get maybe a third lower power consumption. At worst you have double the power consumption, an extra 2kW, and say one more for the AC. That's 30 YEARS before you make up the difference in initial cost. Therein lies the problem with SeaMicro claims. They compare power consumption against hardware two and three generations old, the servers that are going to be replaced. If you actually compare them against new hardware, they're pretty mediocre.

Re:512 Atoms in 10U (0)

Anonymous Coward | more than 3 years ago | (#36879468)

Actually, I believe that part of the benefit to using these many-core atom based solutions is that the atom is designed with power saving capabilities that allow it to throttle down and use fewer watts. Additionally my consolidating power consumption to fewer power supplies there is less overhead (waste), since redundant (standby) power supplies tend to pull the same load as active power supplies in most commercial server designs.

So, if the server is at full load 100% of the time, the benefit is probably negligible. But in theory, the atom solution allows for greater off-hours savings as the processors are allowed to throttle down.

Re:512 Atoms in 10U (1)

wagnerrp (1305589) | more than 3 years ago | (#36879942)

Actually, you would be wrong. One of the things Intel ripped out of the desktop Atoms in the interest of cutting costs is the power management hardware. There is no clock scaling, and little difference between idle and full load power consumption. For a nettop with a single processor plugged into the wall, a couple watts one way or another won't make a noticeable difference. When you have 256 (or 384 on the larger one) all running nearly full consumption all the time, the story is different.

Re:512 Atoms in 10U (1)

TheGratefulNet (143330) | more than 3 years ago | (#36879522)

atom only has a very short time when it will be relevant (some argue its no longer relevant and some even say it never was due to the chipset being the power hog, not the cpu).

seamicro tries to abstract or virtualize the whole hardware and make it 'visible' by all the cpus. sorta kinda.

thing is, atoms still suck as cpus and the big box is still a power and noise pig.

nice try but not really worth the price.

How about 20w xeons? (1)

Pegasus (13291) | more than 3 years ago | (#36880512)

I wonder how would latest sandy bridge based xeons compare to these ... E3-1220L has two cores and 20W tdp, I'm trying to find it for my home server for the next 5 years or so.

Re:How about 20w xeons? (1)

wagnerrp (1305589) | more than 3 years ago | (#36881572)

You're looking at maybe 3x the total performance, and 4-5x the single threaded performance of the Atom, at roughly twice the power consumption.

Re:How about 20w xeons? (0)

Anonymous Coward | more than 3 years ago | (#36882858)

Have you looked at AMD E350 systems (Zacate)? Those are supposed to be about 20W, not sure about cpu comparison. I have some ASRock miniITX boards that are quite nice, for about $110. I'm heard there are four core versions of the E350 coming this fall, could be very interesting.

(oh, I haven't been able to get ESXi running on mine, but others apparently have)

Re:512 Atoms in 10U (1)

serviscope_minor (664417) | more than 3 years ago | (#36881194)

They're using 256 dual core Atoms, for 512 cores at 1.6Ghz.

Well, that's not great from the atom point of view.

Those 2.3Ghz Opterons will do roughly twice the work per core as the Atoms,

That sounds a little low. The clock rate of the opeterons is higher, they are much wider (more superscalar), and have out of order execution. From the old netbook benchmarks, the 1st gen P3 900 (somewhat comparable to the opteron core in terms of features, but older now) runs about as fast as as the 1.6GHz atom. I would expect the opteron cores to go 3-4x as fast, at least.

with eight cores per chip, four chips per box, a 1U box will replace roughly 32 Atoms,

The 6100s come in 12 core varieties at a reasonable price. 48 cores per box, so 1U replaces probably 96 or more atoms.

So how much will this thing cost you?

Well, hard to get comparable specs. The sea micro people don't talk much about RAM. RAM is by far the biggest cost of the 6100s. EUR8500 would (a few months ago) buy you a complete 48 core box with 256G RAM. The full 1/2 TB would cost more like EUR16000. Of course at EUR8500 per box, you get 2TB RAM into the 8U. That would be equivalent to having 4GB per atom, which sounds reasonable.

$900 each for dual port 40Gbps cards

Every year it astonishes me how cheap cool stuff has got.

That adds up to just over $61K, versus $150K for the SeaMicro box of roughly equivalent performance.

Well, your raw numbers are different, but a similar price/performance. I gather that the 2.3GHz 6100s are not the most efficient. At the time I got them, they were the fastest, though 2.5GHz ones are available now. If you invest in the HE versions, it will cost more per chip and probably an extra server or two to get the same performance, but that could easily wipe out the power difference.

They compare power consumption against hardware two and three generations old, the servers that are going to be replaced. If you actually compare them against new hardware, they're pretty mediocre.

They always do. The 48 processor 1U 6100s offer astonishing performance. In terms of price/performance, they are unbeatable. In terms of efficiency, and density they are very competitive. And very cheap.

The intel ones are similar but usually cost 2k more. Given that it's getting towards 6-8K per machine, that pushes it in favour of AMD.

Re:512 Atoms in 10U (1)

wagnerrp (1305589) | more than 3 years ago | (#36881682)

Those 2.3Ghz Opterons will do roughly twice the work per core as the Atoms,

That sounds a little low. The clock rate of the opeterons is higher, they are much wider (more superscalar), and have out of order execution. From the old netbook benchmarks, the 1st gen P3 900 (somewhat comparable to the opteron core in terms of features, but older now) runs about as fast as as the 1.6GHz atom. I would expect the opteron cores to go 3-4x as fast, at least.

Entirely possible. I intentionally chose a safe underestimate of the performance difference.

The full 1/2 TB would cost more like EUR16000. Of course at EUR8500 per box, you get 2TB RAM into the 8U. That would be equivalent to having 4GB per atom, which sounds reasonable.

The SeaMicro box I saw priced for $150K only had 1GB of memory per core, 512GB total. My estimate was for double that, 128GB per quad socket server.

$900 each for dual port 40Gbps cards

Every year it astonishes me how cheap cool stuff has got.

I'm not sure where I read they used infiniband internally, but that seems to be not the case. That's likely just an external interface. Internally, they use some custom toroidal design similar to older supercomputers, which is likely where the bulk of their cost goes, and why they rate so high on very interdependent tasks like mapreduce. However, there is still a lot of free space on that switch, so if 640GBps between the nodes doesn't cut it, another $7200 and you can double it.

I gather that the 2.3GHz 6100s are not the most efficient. At the time I got them, they were the fastest, though 2.5GHz ones are available now. If you invest in the HE versions, it will cost more per chip and probably an extra server or two to get the same performance, but that could easily wipe out the power difference.

Very likely correct. Rather than search out the best performance per watt, I was simply continuing using the same Magny-Cours chip the OP had mentioned.

Re:512 Atoms in 10U (1)

avandesande (143899) | more than 3 years ago | (#36884500)

If your data center has limited power the savings go far beyond the price of the unit.

Re:512 Atoms in 10U (0)

failedlogic (627314) | more than 3 years ago | (#36878892)

"Basically, the article is far too light on details."

So is your comment since you only detail the power consumption. You fail to mention how many Libraries of Congress these systems can process. This is the all important benchmark on Slashdot and if you can't tell us this then I hope you don't get modded informative.

I should be modded informative since I will inevitably inspire someone to respond to this post with the correct answer to "How many Libraries of Congress this system process?". `

Re:512 Atoms in 10U (1)

TheGratefulNet (143330) | more than 3 years ago | (#36879570)

have you tried the intel i3 (i5, i7) cpus? and their chipsets?

I doubted it until I got one. go touch the heatsink on the i3 (any modern chipset) northbridge. (do they still call them NB's? maybe not.) at any rate, its cold to the touch even with the system doing things! totally shocked me.

I can turn the fan OFF on my i5 system. I can run 100% fanless for minutes on end, even half an hour or more if I'm doing light office/web work. even just one gen before (the core2 duos and quads) were so much hotter and these new chips and processors are almost passive coolable!

atom is a dead cpu. somehow, intel (maybe amd, too; but I have not bought amd for years, to be honest) has taken power savings to a new level. the size of the NB heatsink is tiny and it needs no fan. the chassis cooling in a normal i3/5 system needs no fan. only during video rendering does the fan make any real noise. unbelievable (in a good way).

if I was running a datacenter. I'd take a miss on these atoms. too little too late. ok for netbooks, sort of; but nothing else, really.

Re:512 Atoms in 10U (1)

0123456 (636235) | more than 3 years ago | (#36879780)

atom is a dead cpu.

I recently replaced my Atom MythTV server with an i5, and while the i5 is about 5x faster and only uses about 2x the power under load the i5 CPU and motherboard alone cost more than the complete Atom system.

Re:512 Atoms in 10U (1)

wagnerrp (1305589) | more than 3 years ago | (#36880164)

Did the Atom provide sufficient power to run the MythTV server? If it didn't, then however cheap it may have been is irrelevant.

Re:512 Atoms in 10U (1)

0123456 (636235) | more than 3 years ago | (#36880238)

Did the Atom provide sufficient power to run the MythTV server?

It ran fine as an SD PVR for over two years, but it's just too slow for HD transcoding in a reasonable amount of time. It's still running as a Zoneminder server, Zabbix server and game server.

Re:512 Atoms in 10U (1)

Penguinoflight (517245) | more than 3 years ago | (#36880202)

You can't really gauge the efficiency of sandybridge (the current cycle of intel i3/i5/i7 cpus) by their "northbridge" heat. It's true that these chips don't draw so much power, but with sandybridge platform has memory controllers and integrated graphics right in the CPU; you can guess that these two components draw the most power. It's probably still a minor win for power efficiency, but it's not like there's been a huge efficiency breakthrough.

Re:512 Atoms in 10U (1)

0123456 (636235) | more than 3 years ago | (#36880266)

It's probably still a minor win for power efficiency, but it's not like there's been a huge efficiency breakthrough.

I disagree: one reason why I decided to replace my Atom with an i5 is that the Sandy Bridge systems are extremely power-efficient compared to PCs of a few years ago. I haven't measured the power consumption of mine yet, but online benchmarks show i5 systems with integrated graphics idling at half the power usage of my Atom system.

Re:512 Atoms in 10U (1)

timeOday (582209) | more than 3 years ago | (#36880408)

I recently got an i7 4-core MacBook Pro and wonder if I should have got an i5, because this one can hardly do anything without the fan coming on, and is often uncomfortably hot on my lap.

Re:512 Atoms in 10U (1)

bemymonkey (1244086) | more than 3 years ago | (#36880608)

"have you tried the intel i3 (i5, i7) cpus? and their chipsets?

        I doubted it until I got one. go touch the heatsink on the i3 (any modern chipset) northbridge. (do they still call them NB's? maybe not.) at any rate, its cold to the touch even with the system doing things! totally shocked me.

        I can turn the fan OFF on my i5 system. I can run 100% fanless for minutes on end, even half an hour or more if I'm doing light office/web work. even just one gen before (the core2 duos and quads) were so much hotter and these new chips and processors are almost passive coolable!"

Interesting, I assume this is on desktops? The laptop space is more or less the other way around - where Core2Duo Penryn 25W systems would run completely fanless at ~45C, Core iX (both first and second gen, the 35W GPU-on-chip versions, which are more or less equivalent to the older 25W+GPU TDP versions, I suppose) shoot up to 60-70 quickly with the same workload (just idling, reading PDFs and occasionally loading a new web page). It's even worse with CPU+GPU load - the Core2Duo systems top out at 70C in a warm room, while many Core iX systems are at 90+...

Just take a look at systems like the Thinkpad X series... cool and quiet on Core2Duo, hot and loud on Core iX (to the extent of the processor having to go into thermal throttling mode). The Sandy Bridge i3s seem to be alright, but they're so cut down on features that they're not particularly usable.

Re:512 Atoms in 10U (1)

serviscope_minor (664417) | more than 3 years ago | (#36881226)

have you tried the intel i3 (i5, i7) cpus? and their chipsets?

I have an i7 desktop (first gen 975). That thing can chuck out heat on full. I love the i7s because of the really high pre-thread performance. That makes them fantastic for developing stuff on and running small quantities of batch jobs with a quick turnaroud. I also love my AMD compute servers.

I've never tried disconnecting the fan. It's conneded to one of those sealed all in one water cooling units.

Re:512 Atoms in 10U (1)

petermgreen (876956) | more than 3 years ago | (#36881852)

I doubted it until I got one. go touch the heatsink on the i3 (any modern chipset) northbridge. (do they still call them NB's? maybe not.)

There isn't really a northbridge in a LGA1156 or LGA1155 system. The main functions traditionally provided by a northbridge are the memory controller, the high speed IO (usually used for the graphics card though it doesn't technically have to be these days) and the integrated graphics (if present). With LGA1156 and LGA1155 these functions are integrated into the CPU.

The chip you are feeling is probablly the PCH which is essentially the eqivilent of a southbridge. Southbridges always ran much cooler than northbridges.

Re:512 Atoms in 10U (0)

Anonymous Coward | more than 3 years ago | (#36880012)

ARM is the future of this. Try and count how many iPhone or Galaxy S II mainboards you could sit in a 2RU.
They'll slot in like SD cards, with a spring loaded release. They'll get power, rs232, and a fibre connection (PON?) from the 'backplane' at the bottom of the drawer.
4-8 cores and 1-8gb of ram each. They won't be particularly fast on their own, but collectively they'll outpace anything today on less power than your current mobile phone uses.

Re:512 Atoms in 10U (0)

Anonymous Coward | more than 3 years ago | (#36881014)

1. you get 8 socket opteron systems in 1U or 4 socket opteron systems half width so you can fit 2 of them in 1U.
2. 12 cores opterons exist

12*8*10 = 960 opteron cores

so it's 960 opteron cores against 1024 atom cores, uneven battle.
BUT it's not always the cpu which counts!

Tilera and memory bandwidth? (1)

White Flame (1074973) | more than 3 years ago | (#36877800)

Last I looked at Tilera's offerings, their core count to memory controller ratio was really high. It seemed to be really focused on purely streaming data applications, like packet inspection or video conversion.

Anybody know if how Facebook is using them is actually memory-constrained, or it's just low power enough not to matter when 80 quadjillion requests are eventually handled quickly enough, regardless of actual latency?

Also, lack of ARM is disappointing.

Re:Tilera and memory bandwidth? (1)

TheGratefulNet (143330) | more than 3 years ago | (#36879506)

ARM is not i686 compatible.

that's the ONLY real advantage of atom, is that it runs binary code that 'everyone already has'.

ARM only works well if you own source. most data centers do not (not all of it).

Re:Tilera and memory bandwidth? (1)

imroy (755) | more than 3 years ago | (#36879702)

ARM is not i686 compatible.

Neither are Tilera's processors. Being x86-compatible is often unnecessary in the non-PC, non-Windows world. If you're running a web server farm with Linux boxes and server software written in Java, Perl/Python/PHP, or even C/C++, the processor architecture matters very little. And that's the way it should be.

Re:Tilera and memory bandwidth? (0)

Anonymous Coward | more than 3 years ago | (#36881368)

server software written in Java, ... the processor architecture matters very little. And that's the way it should be.

Unfortunately, if you plan to run the Oracle JVM, yes it does matter. Only a few {operating system, architecture} tuples are supported. For example, no {openbsd, sparc}.

Re:Tilera and memory bandwidth? (3, Insightful)

Medievalist (16032) | more than 3 years ago | (#36882792)

Unfortunately, if you plan to run the Oracle JVM, yes [processor architecture] does matter. Only a few {operating system, architecture} tuples are supported. For example, no {openbsd, sparc}.

Thank you! Yet another good reason to avoid Oracle.

Re:Tilera and memory bandwidth? (1)

TeknoHog (164938) | more than 3 years ago | (#36881424)

Exactly. I guess it's news when a big business discovers the world outside x86, where some of us have been living for years.

nice things (1)

luther349 (645380) | more than 3 years ago | (#36877916)

i got one of those tiny plug in servers that draw pretty mutch no power acting like a nas on a usb hdd. it does its job very well.

WOOT.... fp (-1)

Anonymous Coward | more than 3 years ago | (#36877998)

Dim. If *BSD is lube or we seel Out of bed in the all along. *BSD BSD managed to make you are a screaming website Third, you maggot, vomit, shit it a break, if Sux0r status, *BSD

YUO FAiL IT (-1)

Anonymous Coward | more than 3 years ago | (#36878082)

a need to play a BSD box (a PIII centralized distribution. As BOMBSHELL HIT play parties the over a quality blue, :rubber DOG THAT IT IS. IT on baby...don't Yes, I work for maintained that too the accounting may be hurting to foster a gay and a conscious stand numbers. The loss Can no longer be rivalry. While shit-filled, Was in the tea I 'first post' forwards we must from one folder on FreeBSD went out Survival prospects = 36400 FreeBSD FreeBSD went out As little overhead claim that BSD is a declined in market

Price of Quanta's Tilera servers? (1)

White Flame (1074973) | more than 3 years ago | (#36878332)

I've been googling around; their site is just a "Contact us" sort of sales deal, and no other hit seems to mention a price, like can sometimes be found in articles about headlining sales.

Anybody in the know here?

details of the CPUs? (1)

reiisi (1211052) | more than 3 years ago | (#36878826)

That's something else I'd like to know.

It's like, "We don't know anything about pricing yet. But trust us, we'll do it right."

And, then, "We know everything you need to know about the CPU so you don't need to. Trust us. We'll do it right."

Or is it, "If you have to ask about price, you needn't bother." followed by, "If you have to ask about the CPU's instruction set, register count, etc., you needn't bother."

I know, CPU is so passe, now, but I'd still like to know. (Yeah, and money is also passe.)

Re:details of the CPUs? (0)

Anonymous Coward | more than 3 years ago | (#36880254)

Tilera licensed MIPS, so probably either MIPS or similar.

Re:details of the CPUs? (1)

White Flame (1074973) | more than 3 years ago | (#36880892)

Tilera describes their instruction set as "64-bit VLIW with 64-bit instruction bundle", so that doesn't sound quite like stock MIPS if they truly mean multiple independent parallel instructions per 64 bit instruction word. However, MIPS is also desirable for their low-power features, so their licensing might have been for something along those lines.

Re:details of the CPUs? (1)

White Flame (1074973) | more than 3 years ago | (#36881116)

Ah, one of the comments on the article linked to the original paper: http://gigaom2.files.wordpress.com/2011/07/facebook-tilera-whitepaper.pdf [wordpress.com]

This board uses the older 32-bit model CPU, 64 registers, 3-way VLIW, 5-deep pipeline, in-order, 64KB of L2 per core, 24KB of L1, cache coherence with an automatic subscription model and all the L2 cache combining to the effective L3 cache, 2 user-space mesh interconnect lines available between cores, full MMU with up to 64GB physical memory (still 32 bit, so 4GB per process), etc.

Re:Price of Quanta's Tilera servers? (1)

wagnerrp (1305589) | more than 3 years ago | (#36880208)

Last I checked several months ago, the 256 processor (512 core) version was going for $150K.

Discount Christian Louboutin Sale (-1)

Anonymous Coward | more than 3 years ago | (#36878380)

Lighten up! In a pair of yellow Christian Louboutin Shoes [louboutinshoecheap.com] of course! Yellow sandals can instantly bring a smile to your face and a pop of color to your Cheap Louboutin Pumps [louboutinshoecheap.com] . There really is no other color that is more fun to wear during this season because it will really add so much character and personality to your Discount Christian Louboutin Sale [louboutinshoecheap.com] , which will be able to make you stand out and be very approachable.

Cheap Louboutin Shoe [louboutinshoecheap.com] will help create a fun look to your everyday attire. All you really need is a little bit of something bright in your wardrobe to put a Christian Louboutin Booties [louboutinshoecheap.com] when you head out to face the world. So put a smile on your face and some yellow on your Christian Louboutin Slingbacks [louboutinshoecheap.com] ! http://www.louboutinshoecheap.com/ [louboutinshoecheap.com]

Is that why my firefox crashes daily ? (1)

obarthelemy (160321) | more than 3 years ago | (#36878422)

Please Mozilla, go back to real CPUs please. And to fixing bugs, instead of playing with fancy servers.

slashvertisement alert (0)

Anonymous Coward | more than 3 years ago | (#36878768)

(Disc: I interviewed at this company and after interviewing, decided not to work there. Yes, even in this down-economy. Their technical approach and company style just did not pass the smell-test.)

Ok, that out of the way - this is really a not-so-subtle advertisement for the atom webserver company. Why is slash taking so many blatant ads and trying to make them 'tech articles'?

Hate to say it, but atoms had a very short time when they made sense. They simply do not, now, and S.M. is trying to get as much mind-share from it as they can before they lose the little bit of advertising they can bribe from 'partners'.

Oh well... I'm not sure anyone here is really impressed by atoms in a cabinet. the intel i-series of cpus, those can do whatever you want and they are damned low power, too. Atom had such a short life/use cycle and S.M. just was late to the party. Maybe 2 years ago this would have been useful but not really today, sorry.

Cores vs. Processing power? (0)

Anonymous Coward | more than 3 years ago | (#36879026)

Somebody want to explain why not just virtualize the load if processing power isn't needed, just instances of cores?

Not sure what kind of CPU load the main article is about, but I'd think that most of the replies here that suggest using a laptop instead of a server, would benefit from VMs.

Or is there some reason VMs won't work?

Re:Cores vs. Processing power? (1)

White Flame (1074973) | more than 3 years ago | (#36880950)

If your load isn't very computationally expensive, then you can have weaker cores running them "fast enough" at a fraction of the power, rather than a slice of a very fast machine churning through it very quickly while taking lots of power. It's all about performance ratios against power.

But I don't think the product's intent is talking along the same lines as people here, the latter namely being consolidating unused computing potential. Rather, it's more about getting machines taking less power while having the same effective computational capacity.

So run your big threaded app or VM array on N fast cores with X power consumption, vs getting the same (or greater!) overall effective speed on N+M slower cores at X-Y power consumption, as the promise has been for a long time. And systems like web servers and such can make do with tons of requests, each being handled on a slower core. And^2, the speed of the various individual low-power cores is nothing to sneeze at nowadays.

However, this is all pretty newly forged territory, so we'll have to see what this means in the real world at the moment.

Hmm... SGI Was Right About It (1)

PotatoHead (12771) | more than 3 years ago | (#36880526)

Just a bit too early...

If low powered computers are better... (0)

Anonymous Coward | more than 3 years ago | (#36881188)

If low-powered computers are better at sending HTTP responses, imagine what a beowulf cluster of infinite monkeys can do.

Proofreading... (1)

CowardWithAName (679157) | more than 3 years ago | (#36884164)

... is something that's not found st Slashdot.

How Frequently Do You Kill or Leave a Project? (2)

eldavojohn (898314) | more than 3 years ago | (#36887910)

It looks like you've been involved in many projects. I've got about 10 different side projects (outside of work) going on at any given time in several different realms. How often do you decide it's time to end a project so that you can focus on a better project? Have any projects that you devoted a lot of time to result in nothing or have all come to fruition in one way or another? What is your criteria for this?
Load More Comments
Slashdot Login

Need an Account?

Forgot your password?