Atom Processors Set New Record For Power-Efficient Sorting 92
schliz writes "German researchers have set a new record for energy efficient data sorting with a system based on netbook processors and Solid State Disks. The system, dubbed EcoSort, more than tripled the power efficiency of former record holders, leading one of its developers to claim: 'In the long run, many small, power-efficient and cooperating systems are going to replace the so far used, heavy weighted ones.' Records were defined by 'Sort Benchmark,' which was created by missing Microsoft scientist Jim Gray and was now managed by representatives of companies like Hewlett-Packard and Microsoft."
Comment removed (Score:5, Informative)
Re: (Score:1, Redundant)
Actually this is less about bills and more about battery life.
This is the Achille's heel when it comes to mobile computing and until more of these breakthroughs are made it will be the one, most important, limiting factor.
Re: (Score:2)
Re: (Score:1, Interesting)
I'm sure I could phrase this better... I've got to admit I don't know much about low-level hard disk theory. But I do recall seeing something regarding researchers trying to extend the life of SSDs, since their life is not nearly as long as s
Re:This is the way we are headed (Score:4, Insightful)
Short answer: No.
Long answer: They sort of used to be, but nowadays the lifecycles and capacities are large enough such that you could keep the SSD's interface saturated with writes for 5-10 years straight before you start to encroach on their conservatively rated write cycle life expectancy.
How are you sure they'll last that long? (Score:5, Insightful)
It's easy for some marketing fools to say, "Oh, for sure, it'll last 5 to 10 years." It's easy for them to print those claims on the product packaging, too. But marketing claims don't, of course, have any real impact on the lifespan of a product.
We heard the same claims for CD-Rs years back. They'd last 99 years, we'd often hear. Now, less than 10 years later, people who backed up data onto CD-Rs are running into problems. Even when storing the burned CD-Rs properly, they have nevertheless developed unrecoverable read errors because they've degraded many times faster than expected.
Frankly, we can't say that these SSD drives will last 5-10 years straight, while saturated, especially while they really haven't been around for that long. Unless you've actually taken a drive and had it perform writes continuously for a decade, and can demonstratively provide that the drives will last that long before performance degrades, we have to assume the worst.
Re: (Score:1)
So lets assume it takes 3 years of proper use(massive torrenting of porn and deleting it afterwards) to reach the performance of a normal slow HD?
Re: (Score:2)
Kudos, good sir.
Re: (Score:2, Insightful)
I think we can safely say that you can saturate it for a couple years, as I imagine someone has done that and not had any issues.
Though I haven't seen the data, I think if someone consistently showed SSDs dying at a year of saturation (which is far more than you will usually have) it would make news.
Re: (Score:2)
The bigger problem is write fragmentation. I have a one year old SSD (Intel X25-M, first generation) with about 15 gigabytes of writes per day and earlier this year I had to do a secure wipe on it to reset the fragmentation tables. With TRIM support this problem should go away.
I would really like to know when RAID controllers start supporting TRIM though, and when it'll be possible to hook up a readzilla/logzilla SSD like Sun has to a server to serve as a large second level cache and non-volatile write log.
Re: (Score:1)
Frankly, we can't say that these SSD drives will last 5-10 years straight
Frankly, we can't say anything about many things. If the risk of a gadget failing withing 5-10 years is unbearable, buy insurance against it.
Re: (Score:1)
Frankly, we can say anything about anything... and some of us would mean everything we say.
That does not, however, mean we know what the hell we are talking about.
Re: (Score:2)
This is probably true, but it will most likely outlast most hard disks.
But it is also true that many hard disks die after a couple of years average now, sometimes less...especially those that come in the pre-built retail computer systems. In some cases if you try to use the hard disk manufacturers warranty directly, they will tell you no and suggest you call system manufacturer that sold it to you (makes me wonder if the drives are refurbs or seconds, or perhaps they were spec'd out to be cheaper models),
Re: (Score:1)
Re: (Score:2)
> Unless you've actually taken a drive and had it perform writes continuously for a decade
Not true at all. That's why people does statistical modeling. The failure rate usually follows a exponential distribution, so you could calibrate their behavior in a fraction of time by analyzing a big set of disks.
Re: (Score:2)
Re: (Score:1, Offtopic)
First off, I apologize for the off-topic post in advance. I couldn't find any contact links to report this.
The Firehose has been Spam botted. Only 3 of the current 20 or so submissions are actually story submissions, the rest are spam/adverts for everything from acai berry shit to resorts in Goa. Spam posts about every 3 mins now...
And whats with all the wall-of-text repeat troll posts? They are in pretty much every story thread these days.
Could someone that knows how, or who, to contact about this (the spa
Re: (Score:2)
I suggest a netbook. You are limited to a 12.1 inch screen, but even the cheapest Asus will run for 4-5 hours of moderate use. Several models have 8+ hour ratings.
Re: (Score:2)
noise level (I love watching videos while my wife sleeps)
You might just wake up the wife instead of relying on videos...
heat (I hate it when the laptop is hot and I actually have it on my laps ...)
And that's a bad thing how?
autonomy (I travel quite a bit in trains and planes, and my 2 hour autonomy laptop is not so great ...)
I hope those trains and planes are often rather empty...
Re: (Score:2)
The system is based on a Zotac IONITX-A board, equipped with an Atom 330. This processor consumes more than three times the power of an N270 (8 W TDP) but supports two cores and four hardware threads. The main advantage of this system is that its nVidia Ion chipset provides four SATA ports that can handle the SSD transfers at full speed. Moreover, it allows two DIMMs for a total of 4 GiB of RAM. The 64 bit logical address space is less prone to fragmentation, which we experienced on the 32-bit Atom N270.
Which really shows that they got a 3 times improvement by going with sub-optimal choices because that's all that was available. If there existed a two or 4 core N270 64 bit that had the necessary RAM and SATA interfaces the results would have been even better. Just think if there
Re: (Score:2, Funny)
no no no — you're supposed to say “Can you imagine...”
Re: (Score:1)
Actually I have seen pictures of server farms using massive amounts of ITX boards.
One of the added benefits is to have a RAID style set-up aka RAIN : 'Redundant Array of Inexpensive Nodes' where you just have added hardware you might not need now, but since they are cheap, you just add more and power wise it also won't hurt much.
I have to still fight with 'old-timers' who still think in 'big and powerful monolithic' systems.
If I can get 2x the power with 1/2 the costs, what will the customer take?
Re: (Score:3, Insightful)
Re: (Score:2, Interesting)
True, though the space used by an ITX is smaller then those for your typical blade.
You can get about 2-3 in per u.
I also think you have to calculate the power needed. Some of the systems I have seen are designed for the worst-case, yet that hardly ever happens. But the system has to have all the bling and whatnot.
Or, if you need the power, the system has been nerfed because each node costs so much.
Also not to dismiss is the cooling requirements.
I have seen nodes you could use to keep your coffee/tea warm (o
Re: (Score:3, Interesting)
Not much in the article (Score:5, Informative)
Good to see that Jim's work lives on...meanwhile, this is about all you get in the article:
"EcoSort set records in the Joule category, which measured the amount of energy required to sort either 10GB, 100GB or 1TB of records.
It reached a maximum efficiency of 36,400 records sorted per joule for 100GB of data, using an Intel Atom 330 processor, 4GB of RAM, and four 256GB SSDs by flash vendor Super Talent Technology.
In 2009, a team from the University of Melbourne had the 100GB record of 11,600 records sorted per joule using the OzSort system, which comprised a 2.6GHz AMD processor, 4GB of RAM, seven 160GB 7200 RPM SATA hard disks and a Linux operating System."
Sure, this is the way things are going, but until prices come down we won't be seeing SSDs replacing HDDs; work fine for the desktop, tho'
Re: (Score:2, Informative)
More details are here [sortbenchmark.org]. Looks like it's a tweaked merge-sort.
Re: (Score:2)
until prices come down we won't be seeing SSDs replacing HDDs
SSDs wont be as cheap per GB as HDDs for years and years, but that doesn't mean SSDs dont have their application already today. A 80 GB SSD is already quite affordable and holds enough data to be useful for a lot of people. And of course there's always the option to put your large data on HDDs (photos, videos, music, porn, whatever), and run your OS and applications from an SSD to get the benefit of the increased access times.
Re: (Score:2)
We won't be seeing SSDs replacing HDDs until people start calculating differently: (Total power consumption costs + total cooling costs + price) / years until failure or replacement. And let's face it - 128GB of storage is enough for a lot of servers.
Re: (Score:2)
why is the Via C7 not more popular? (Score:5, Interesting)
The Via C7/Nano seems to be a great chip for a home/small office server, what with its built-in AES encryption making it faster than even a high end Xeon without hardware acceleration. My current setup consists of 2*WD SE16 hard drives, APC UPS, 80+ Corsair PSU, PC2500e Nano mobo with 1GB, and a couple of 80mm case fans, together running under 50W idle, and only 7W more at full CPU load. If I were to replace the Corsair with a fanless PSU good up to 80-120W I might get an extra 5-10% efficiency; I could wipe out the case fans probably with no problem (2-3W, say), especially if I replaced hard with spinning solid state storage, and that of course would shave off around 15W. Substitute a large fanless heatsink for another W (or just get a fanless motherboard/CPU in the first place). But even as-is, it's a good improvement on my previous regular desktop CPU-based setup.
For something which is on 24 hours a day, going several months between reboots and stressed only in the IO and encryption departments, I see no reason to use a full-power desktop processor. So, what problems have you guys encountered which has meant you haven't ended up with this option?
Re: (Score:1)
Guess the Atom is more energy efficient and faster than C7 (1Ghz is quite low ). Let alone now there are dual core atom chips, requiring only 4-5 watts more than the single core ones.
Re: (Score:2)
Wait, what? I was talking about a specific application in the general topic of power-efficient CPUs. I've yet to see a comparison of current high end Intel vs Via, e.g. Atom D510 vs Via Nano. If you genuinely need to sort all day, you probably have the intelligence and resources to prepare yourself, so you might be better off building a hardware sorting algorithm in an FPGA. You're unlikely to cluster low-power CPUs on off-the-shelf motherboards because your interconnect will be shit.
Also, what is the 1GHz
Re: (Score:1)
Obviously they're looking at stock hardware. Furthermore, all CPUs are about as optimized as anything could be for sorting. Trying to roll your own in hardware isn't going to help a lot when the primary bottlenecks are memory and disk.
Re: (Score:2)
Is the built-in AES encryption useful on Linux installations? E.g. if I scp files, will the encryption/decryption get offloaded to hardware?
Re:why is the Via C7 not more popular? (Score:4, Informative)
Yes. The bottleneck for ssh probably becomes the HMAC - this thread is enlightening [mindrot.org]. SHA HMAC is afaict considered difficult to accelerate with Via's implementation, though a patch may exist. With a stock (AES-accelerated) build, software MD5 is quicker.
Re: (Score:2)
Sort of. You have to install the VIA Java Cryptography Service Provider to use the hardware acceleration, and I have the distinct feeling that standard SSH implementations won't be able to use that unless you patch them accordingly.
Re:why is the Via C7 not more popular? (Score:5, Informative)
Erm, no, for developers there have been Linux kernel crypto modules supporting the Via Padlock [logix.cz] included since 2.6.11, and if you don't want them, you can always use the crypto instructions directly. The Java is just an API option provided by Via.
Re: (Score:1)
Oops. Didn't know that, thanks for clearing it up. I think I'd rather make my next home server CPU a C7 instead of an Atom then!
Re:why is the Via C7 not more popular? (Score:4, Interesting)
What are you using AES encryption for, hard disk encryption? If so, this is a little unusual for a server, which are normally found in secure facilities, but make sense for a home perhaps.
I'm thinking a home NAS isn't something one would want a common house-thief to walk away with. But TFA article talks about sorting, and not NAS work, hence my request for clarity. I'm curious what your application, and OS is. Your setup is certainly interesting.
Re:why is the Via C7 not more popular? (Score:4, Informative)
Sorry - although I mentioned that the Via's great for a home/office encrypted NAS, I perhaps wasn't clear that this is precisely the application I was talking about. I was just expanding the discussion on power-efficient CPU applications, and implying that, when considering energy efficiency, a low power CPU with dedicated circuitry for popular complex operations might be the way forward.
Re: (Score:2)
Thank you very much for the clarification, and the original information! I never really considered a VIA NAS PC, but yeah, disk encryption makes much sense for such an application.
And I had no idea the VIA CPU offered such disk-encryption in-chip performance. As someone who really enjoys Ubuntu full-disk-encryption on notebooks, and also as someone who has considered this type of CPU for home/soho NAS-use but hasn't gotten so deep yet... Thank you very much for the low-power, home/soho security CPU tip! (I
Re: (Score:2)
Many latest Intel processors also include AES acceleration: http://en.wikipedia.org/wiki/AES_instruction_set [wikipedia.org] (too bad Intel, as is usual with them, castrates low-end too excessively)
While not really beneficial over Via in NAS scenarios, I guess you would be happy if your next laptop had such CPU...
Re: (Score:1)
best steer clear of SiS then too... i have an intel mini-itx board with a crappy sis chip which refuses to work properly in linux, for the rest it is quite OK.
but besides, X on server is for wimps :P real men SSH into their boxes and just bash their way to victory!
Re: (Score:1)
For the consumer its down to whats cheaper and what have they heard about, Via never ran big advertisments like intel has done for atom in different mags, so lack of knowledge of the product leads to no comsumer demant.
While the c7/nano compares favoritable to the atom, they have failed to keep up and offer a dual core version, and
Re: (Score:2)
Re:why is the Via C7 not more popular? (Score:5, Insightful)
Re: (Score:2)
Erm, you're comparing apples and orchards.
45-50W is including UPS, PSU and everything spinning (and miscellaneous peripherals omitted from the above list, e.g. second Ethernet PCI, floppy, DVD drive, external modem), and with a discussion on how this is far from the lowest power solution and how I could improve things. The drives spin using around 8W each, for example.
The BeagleBoard draws 50mW at 5V DC with nothing connected, and would be hopeless for the NAS (and many other) applications described because
Re: (Score:2)
and now have an HP 50G calculator running off 4 AA batteries with a faster ARM CPU. I do appreciate the architecture and power efficiency of ARM designs.
Are hp still emulating the saturn processor on their arm devices? while I understand with that model they are aiming for backwards compatibility. Attaching an arm clocked at 75mhz to a measly 512kb of ram and a 131x80 pixel screen seems ridiculous.
I know it's just a calculator and there to get the job done, but native code is nice and quick and using less cpu power lets the batteries last longer.
The TI nspire seems to be the only graphic calculator out there with a modern design.
Re: (Score:2)
Are hp still emulating the saturn processor on their arm devices?
Yes, although some operations have been accelerated with a native rewrite. But the emulation allows a very mature product to be executed without the cost and bugs inevitable with a complete port. People don't use calculators for their horsepower, but for usability and portability. The HP RPN + CAS are very usable.
FWIW, I very quickly gave up learning Saturn assembler when I still had a real Saturn device :-). Today, there are well-known ways of escaping to native ARM.
Re: (Score:1)
But you forget that most of those Watts is used to cool the CPU, and other stuff. A quad core ARM CPU of the newest generation would not need a cooling system, which would mean another big amounts of watts not drawn. Heck, The fact that the quad would draw a lot less than the 5-20watt slow x86 processor would give it a nice advantage.
Re: (Score:2)
For CPU cooling, I have replaced the stock PC2500e fan with a 40mm Rasurbo specced at 0.6W max, which I've then further slowed down. One of the dedicated case fans is used to cool the drives (quite effectively, so it'd be there regardless of CPU), and the other may be extraneous, as the PSU draws out air. It is left over from when I had a higher power board in there, particularly because I was considering a Zalman ZM-NB47J passive cooler for the C7 CPU and then keeping that case fan to ensure good airflow o
Re: (Score:2)
Many people seem to run completely motionless Via C7 boxes, including by modding this motherboard - although more likely is that you'd start off buying a fanless board. They're commonly available clocked at 1GHz, though 1.2-1.5GHz fanless seems possible.
That was pretty accurate about three years ago when I built a fanless mythtv box. Still running great.
Now a days they're all 1.2GHz on a designed to be fanless, off the shelf board. ATOM based boards claim to have a higher marketing speed. No idea if they actually crunch numbers any faster, first stage in the proc might be a /2 flipflop for all I know.
http://www.mini-itx.com/store/?c=2 [mini-itx.com]
The reason for slow speed growth in that market sector is lack of interest... I play full screen video on my 3 year old m
Re: (Score:2)
50W is a little high but not altogether bad for a system with multiple spinning disks and fans. The Beagleboard is an embedded system. Even with the required peripherals it doesn't hold a candle performance-wise to an Atom/Nano-based server.
Re: (Score:2)
Re: (Score:2)
Wotcha. I'm glad the thread's received quite a bit of interest!
(1) Why "should" disks be spun down on an idle server? It's not like CPU which can pretty much instantly switch between power modes: an HDD takes time to spin up and spin down, and its life is decreased in doing so. You may decide to spin down drives, but only after a sufficient period of idling. If you really want to save power, an idle server "should" be asleep with WakeOnLan enabled.
(2) The BeagleBoard has 128MB of what looks to be on-chip me
Re: (Score:2)
I did also considering a bone stock 200$ eeepc 900 series draws only 36 watts under full load. Hook that up to an external monitor and keyboard and you have something much more efficient for web browsing use. Toss in an external hard disk and you're all set to go.
Re: (Score:2)
The EeePc's 36W full load is quite similar to the C7's 55W, recalling that the latter has two 3.5" drives and various fans rather than SSD. If a casual browsing machine were what's required, they'd both be fit for purpose following de-moving-part-ification of the C7.
Others have mentioned laptops (including the Mac Mini, which is just half a laptop) as requiring less power. Well, yes - the aim isn't to find the lowest power machine with storage and net connectivity, as then my 4 AA cell powered ARM-based HP
Re: (Score:2)
Yes, I noticed that one straight after posting. Dammit :-).
Re: (Score:2)
Re: (Score:2)
Which motherboard, drives, PSU? Is everything spinning?
I'm not surprised to hear that idle performance is excellent on a modern desktop CPU, and I've heard similar achievements with a single desktop HDD. The 45mm Core 2 Duos also idle very efficiently, but shoot up under load with no hardware acceleration to mitigate.
I know that I'm unlikely to get readings much below current unless I switch to a PicoPSU or similar, because inefficiencies in a desktop PSU shoot right up at very low power output.
Re: (Score:2)
Re: (Score:2)
There is absolutely no way that you're reading 5W at the wall with one Core 2 Duo plus two spinning drives, sorry. If one of these is a 7200rpm+ desktop drive, 5W+ will be needed to spin one drive. I'm running 4 year old but stably performing WD SE16 drives, and I know I could do better, but not this well.
Re: (Score:2)
Good points, and an apples-to-apples comparison is always difficult. While billed as a desktop, many of the Mini components are parts designed for mobile systems. Internal hard disk is 2.5", as is the external USB drive, and both spin down after a long enough period with no activity. Unused hardware components get powered off. The Core 2 Duo allows parts of the on-chip hardware to be shut down as well. The OS does as much with clock rates and voltage as it can, and does so aggressively. Some of Apple's lit
North bridge, not so much... (Score:2, Interesting)
Of course you set records, when most of your CPU actually sits in your north bridge. Yes. That thing with the large heat sink and fan, is the north bridge. Not the CPU. The CPU is that smaller chip that you thought were the NB.
It’s a fraud. Nothing else. A trick to hide their failure to get even in the same magnitude as ARM.
Re: (Score:2, Interesting)
Here here!
If any one attempted this test with a Arm Cortex 9A with the full 4 cores, this would be blown out of the water easily.
System power is what they measure (Score:4, Informative)
Re: (Score:1)
The newer Pineview Atoms do combine the Northbridge and CPU. Again this is nothing speical or any trickery, AMD has been doing this for years. You would again be wrong in your description if it was a Pineview in that t
Re: (Score:1)
Ok, a windows tool on a unix platform? Woot?
But seriously when you move on.... does that mean that we could actually have tested this with a quad core ARM? I think we have a winner.
Meme built into the summary... (Score:2)
...leading one of its developers to claim: 'In the long run, many small, power-efficient and cooperating systems are going to replace the so far used, heavy weighted ones.'
They're imagining the Beowulf clusters for us...
Re: (Score:2)
Dangerous Metric (Score:1)
Records / joule might be interesting to some people, but it is a dangerous metric overall.
Why?
Because it ignores time.
It is accounted for somewhat in that there will be some power draw for spinning disks, or leakage; but all-in-all not good.
Personally I hope this catches on (Score:2)
Suspicious (Score:2)
Jim Gray, Tim Bray... hm. Has anyone ever seen them in the same room together ?
some data farms exploring cell-phone netbook chips (Score:2)
Re: (Score:2)
But you'll have less density in terms of computing power, which means more racks and more floor space. I think the trend is toward higher power density. Once you add real estate costs, it is cheaper to run everything on fewer high powered CPUs than many Atoms although it may be less power efficient.
Re: (Score:1)
Not most energy efficient sorting (Score:1)