Carpet Bomberz Inc.

Category: data center

  • Facebook: No ‘definite plans’ to ARM data centers • The Register

    Image representing Facebook as depicted in Cru...
    Image via CrunchBase

    Clearly, ARM and Tilera are a potential threat to Intel’s server business. But it should be noted that even Google has called for caution when it comes to massively multicore systems. In a paper published in IEEE Micro last year, Google senior vice president of operations Urs Hölzle said that chips that spread workloads across more energy-efficient but slower cores may not be preferable to processors with faster but power-hungry cores.

    “So why doesn’t everyone want wimpy-core systems?” Hölzle writes. “Because in many corners of the real world, they’re prohibited by law – Amdahl’s law.

    via Facebook: No ‘definite plans’ to ARM data centers • The Register.

    The explanation given here by Google’s top systems person is that latency versus parallel processes overhead. Which means if you have to do all the steps in order, with a very low level of parallel tasks that results in much higher performance. And that is the measure that all the users of your service will judge you by. Making things massively parallel might provide the same level of response, but at a lower energy cost. However the complications due to communication and processing overhead to assemble all the data and send it over the wire will offset any advantage in power efficiency. In other words, everything takes longer and latency increases, and the users will deem your service to be slow and unresponsive. That’s the dilemna of Amdahl’s Law, the point of diminishing returns when adopting parallel computer architectures.

    Now compare this to something say we know much more concretely, like the Airline Industry. As the cost of tickets came down, the attempt to cut costs went up. Schedules for landings and gate assignments got more complicated and service levels have suffered terribly. No one is really all that happy about the service they get, even from the best airline currently operating. So maybe Amdahl’s Law doesn’t apply when there’s a false ceiling placed on what is acceptable in terms of the latency of a ‘system’. If airlines are not on time, but you still make your connection 99% of the time, who will complain? So by way of comparison there is a middle ground that may be achieved where more parallelizing of compute tasks will lower the energy required by a data center. It will require greater latency, and a worse experience for the users. But if everyone suffers equally from this and the service is not great but adequate, then the company will be able to cut costs through implementing more parallel processors in their data centers.

    I think Tilera holds a special attraction potentially for Facebook. Especially since Quanta their hardware assembler of choice is already putting together computers with the Tilera chip for customers now. It seems like this chain of associations might prove a way for Facebook to test the waters on a scale large enough to figure out the cost/benefits of massively parallel cpus in the data center. Maybe it will take another build out of a new data center to get there, but it will happen no doubt eventually.

    Related Articles

  • Data hand tools – O’Reilly Radar

    A Shebang, also Hashbang or Sharp bang. This i...
    Image via Wikipedia

    Whenever you need to work with data, don’t overlook the Unix “hand tools.” Sure, everything I’ve done here could be done with Excel or some other fancy tool like R or Mathematica. Those tools are all great, but if your data is living in the cloud, using these tools is possible, but painful. Yes, we have remote desktops, but remote desktops across the Internet, even with modern high-speed networking, are far from comfortable. Your problem may be too large to use the hand tools for final analysis, but they’re great for initial explorations. Once you get used to working on the Unix command line, you’ll find that it’s often faster than the alternatives. And the more you use these tools, the more fluent you’ll become.

    via Data hand tools – O’Reilly Radar.

    This is a great remedial refresher on the Unix commandline and for me kind of reinforces an idea I’ve had that when it comes to computing We Live Like Kings. What? How is that possible, well think about what you are trying to accomplish and finding the least complicated quickest way to that point is a dying art. More often one is forced to follow or highly encouraged to set out on a journey with very well defined protocols/rituals included. You must use the APIs, the tools, the methods as specified by your group. Things falling outside that orthodoxy are frowned upon no matter what the speed and accuracy of the result. So doing it quick and dirty using some Shell scripting and utilities is going to be embarrassing for those unfamiliar with those same tools.

    My experience doing this involved a very low end attempt to split Web access logs into nice neat bits that began an ended on certain dates. I used grep, split, and a bunch of binaries I borrowed for doing log analysis and formatting the output into a web report. Overall it didn’t take much time, and required very little downloading, uploading,uncompressing,etc. It was all commandline based with all the output dumped to a directory on the same machine. I probably spent 20 minutes every Sunday running these by hand (as I’m not a cronjob master much less an atjob master). And none of the work I did was mission critical other than being a barometer of how much use the websites were getting from the users. I realize now I could have had the whole works automated with variables setup in the shell script to accommodate running on different days of the week, time changes, etc. But editing the scripts by hand in vi editor only made me quicker and more proficient in vi (which I still gravitate towards using even now).

    And as low end as my needs were and how little experience I had initially using these tools, I am grateful for the time I spent doing it. I feel so much more comfortable knowing I can figure out how to do these tasks on my own, pipe outputs into inputs for other utilities and get useful results. I think I understand it though I’m not a programmer, and couldn’t really leverage higher level things like data structures to get work done, no. I’m a brute force kind of guy and given how fast the CPUs are running, a few ugly, inefficient recursions isn’t going to kill me or my reputation. So here’s to Mike Loukides article and how much it reminds me of what I like about Unix.

  • Quanta crams 512 cores into pizza box server • The Register

    Image representing Tilera as depicted in Crunc...
    Image via CrunchBase

    Two of these boards are placed side-by-side in the chassis and stacked two high, for a total of eight server nodes. Eight nodes at 64 cores each gives you 512 total cores in a 2U chassis. The server boards slide out on individual trays and share two 1,100 watt power supplies that are stacked on top of each other and that are put in the center of the chassis. Each node has three SATA II ports and can have three 2.5-inch drives allocated to it; the chassis holds two dozen drives, mounted in the front and hot pluggable.

    via Quanta crams 512 cores into pizza box server • The Register.

    Amazing how power efficient Tilera has made it’s shipping products as Quanta has jammed 512 cores into a 2 Rack Unit high box. Roughly this is 20% the size of the SeaMicro SM-10000 based on Intel Atom cpus. Now that there’s a shipping product, I would like to see benchmarks or comparisons made on similar workloads using both sets of hardware. Numerically speaking it will be an apples-to-apples comparison. But each of these products is unique and are going to be difficult to judge in the coming year.

    First off, Intel Atom is an x86 compatible low power chip that helped launch the Asus/Acer netbook revolution (which until the iPad killed it was a big deal). However Quanta in order to get higher density on its hardware has chosen a different CPU than the Intel Atom (as used by SeaMicro). Instead Quanta is the primary customer for a new innovated chip company we have covered on carpetbomberz.com previously: Tilera. For those who have not been following the press releases from the company Tilera is a spin-off of an MIT research project in chip-scale networking. The idea was to create very simplified systems on a chip (whole computers scaled down to single chip) and then network them together all the same slice of silicon die. The speeds would be faster due to most of the physical interfaces and buses being contained directly on the chip circuits instead of externally on the computer’s motherboard. The promise of the Tilera chip is you can scaled up on the silicon wafer as opposed to the racks and racks of equipment within the datacenter. Performance of the Tilera chip has been somewhat a secret, no benchmarks or real comparisons to commercially shipping CPUs have been performed. But the feeling generally is any single core within a Tilera chip should be about as capable as the processor in your smartphone, and every bit as power efficient. Tilera has been planning to scale up to 100 cpus eventually within one single processor die and appears to have scaled up to 64 on its most recent research chips (far from being commercially produced at this point.)

    I suspect both SeaMicro and Quanta will have their own custom OSes which run as a central supervisor allowing the administrators to install and sets up instances of their  favorite workhorse OSes. Each OS instance will be doled out to an available CPU core and then be linked up to a virtual network and virtual storage interface. Boom! You got a web server, file server, rendering station, streaming server, whatever you need in one fell swoop. And it is all bound together with two 1,100 watt power supplies in each 2 Rack Unit sized box. I don’t know how that compares to the SeaMicro power supply, but I imagine it is likely smaller per core than the SM-10000. Which can only mean in the war for data power efficiency Quanta might deliver to market a huge shot across the bow of SeaMicro. All I can say is let the games begin, let the market determine the winner.

    Related Articles

  • Microsoft Research Watch: AI, NoSQL and Microsoft’s Big Data Future

    Image representing Microsoft as depicted in Cr...
    Image via CrunchBase

    Probase is a Microsoft Research project described as an “ongoing project that focuses on knowledge acquisition and knowledge serving.” Its primary goal is to “enable machines to understand human behavior and human communication.” It can be compared to  Cyc, DBpedia or Freebase in that it is attempting to compile a massive collection of structured data that can be used to power artificial intelligence applications.

    via Microsoft Research Watch: AI, NoSQL and Microsoft’s Big Data Future – ReadWriteCloud.

    Who knew Microsoft was so interested in things only IBM Research’s Watson could demonstrate? AI (artificial intelligence) seems to be targeted at Bing search engine results. And in order to back this all up, they have to ditch their huge commitment to Microsoft SQL Server and go for a NoSQL database in order to hold all the unstructured data. This seems like a huge shift away from desktop and data center applications and something much more oriented to a cloud computing application where collected data is money in the bank. This is best expressed in the example given in the story of Google vs. Facebook. Google may collect data, but it is really delivering ads to eyeballs. Whereas Facebook is just collecting the data and sharing that to the highest bidder. Seems like Microsoft is going the Facebook route of wanting to collect and own the data rather than merely hosting other people’s data (like Google and Yahoo).

    Related Articles

  • Calxeda boasts of 5 watt ARM server node • The Register

    Calxeda is not going to make and sell servers, but rather make chips and reference machines that it hopes other server makers will pick up and sell in their product lines. The company hopes to start sampling its first ARM chips and reference servers later this year. The first reference machine has 120 server nodes in a 2U rack-mounted format, and the fabric linking the nodes together internally can be extended to interconnect multiple enclosures together.

    via Calxeda boasts of 5 watt ARM server node • The Register.

    SeaMicro and now Calxeda are going gangbusters for the ultra dense low power server market. Unlike SeaMicro, Calxeda wants to create reference designs it licenses to manufacturers who will build machines with 120 cores in a 2 Unit rack. SeaMicro’s record right now is 512 cores per 10U rack  or roughly 102+ cores in a 2 Unit rack. The difference is the SeaMicro product uses an Intel low power Atom cpu,  whereas Calxeda is using a processor used more often in smart phones and tablet computers. SeaMicro has hinted they are not wedded to the Intel Architecture, but they are more interested in shipping real live product than coming up with generic designs others can license. In the long run it’s entirely possible SeaMicro may switch to a different CPU, they have indicated previously they have designed their servers with flexibility enough to swap out the processor to any other CPU if necessary. It would be really cool to see an apples-to-apples comparison of a SeaMicro server using first Intel CPUs versus ARM-based CPUs.

    Related Articles

  • SeaMicro drops 64-bit Atom bomb server • The Register

    Image representing SeaMicro as depicted in Cru...
    Image via CrunchBase

    The base configuration of the original SM10000 came with 512 cores, 1 TB of memory, and a few disks; it was available at the end of July last year and cost $139,000. The new SM10000-64 uses the N570 processors, for a total of 256 chips but 512 cores, the same 1 TB of memory, eight 500 GB disks, and eight Gigabit Ethernet uplinks, for $148,000. Because there are half as many chipsets on the new box compared to the old one, it burns about 18 percent less power, too, when configured and doing real work.

    via SeaMicro drops 64-bit Atom bomb server • The Register.

    I don’t want to claim that Seamicro is taking a page out of the Apple playbook, but keeping your name in the Technology News press is always a good thing. I have to say it is a blistering turnaround time to release a second system board for the SM10000 server so quickly. And knowing they do have some sales to back up the need for further development makes me thing this company really could make a  go of it. 512 CPU cores in a 10U rack is still a record of some sort and I hope to see one day Seamicro publish some white papers and testimonials from their current customers to see what killer application this machine has in the data center.

    Related Articles

  • Chip upstart Tilera in the news

    Diagram Of A Partial Mesh Network
    Diagram Of A Partial Mesh Network

    As of early 2010, Tilera had over 50 design wins for the use of its SoCs in future networking and security appliances, which was followed up by two server wins with Quanta and SGI. The company has had a dozen more design wins since then and now claims to have over 150 customers who have bought prototypes for testing or chips to put into products.

    via Chip upstart Tilera lines up $45m in funding • The Register.

    There’s not been a lot of news about Tilera most recently, but they are still selling products, raising funds through private investments. Their product road map is showing great promise as well. I want to see more of their shipping product get tested in the online technology website arena. I don’t care if Infoworld, Network World, Tom’s Hardware or Anandtech does it. Whether it’s security devices or actual multi-core servers it would be cool to see Tilera compared even if it was an apples and oranges type of test. On paper it appears the mesh network of Tilera’s multi-core cpus is designed to set it apart from any other product currently available on the market. Similarly the ease of accessing the cores through the mesh network is meant to make the use of a single system image much easier as it is distributed across all the cores almost invisibly. In a word Tilera and its next closest competitor SeaMicro are cloud computing in a single solitary box.

    Cloud computing for those who don’t know is an attempt to create a utility like the water system or electrical system in the town where you live. The utility has excess capacity, and what it doesn’t use it sells off to connected utility systems. So you always will have enough power to cover your immediate needs with a little in reserve for emergencies. On the days where people don’t use as much electricity you cut back on production a little or sell off the excess to someone who needs it. Now imagine that electricity is computer cycles doing additions, subtractions or longer form mathematical analysis all in parallel and scaling out to extra computer cores as needed depending on the workload. Amazon has a service they sell like this already, Microsoft too. You sign up to use their ‘compute cloud’ and load your applications, your data and just start crunching away while the meter runs. You get billed based on how much of the computing resource you used.

    Nowadays, unfortunately, in data centers you got single purpose servers doing one thing, sitting idle most of the time. This has been a going concern so much so that a whole industry has cropped up of splitting those machines into thinner slices with software like VMWare. Those little slivers of a real computer then take up all the idle time of that once single purpose machine and occupy a lot more of its resources. But you still have that full-sized, hog of an old desktop tower now sitting in a 19 inch rack, generating heat and sucking up too much power. Now it’s time to scale down the computer again and that’s where Tilera comes in with it’s multi-core, low power, mesh-networked cpus. And investment partners are rolling in as a result of the promise for this new approach!

    Numerous potential customers, venture capital outfits, and even fabrication partners are jumping in to provide a round of funding that wasn’t even really being solicited by the company. Tilera just had people falling all over themselves writing checks to get a piece of the pie before things take off. It’s a good sign in these stagnant times for startup companies. And hopefully this will buy more time for the roadmap to future cpus from the company hopefully scaling up to the 200 core cpu that would be peak achievement in this quest for high performance, low-power computing.

    Related Articles

  • Next-Gen SandForce Controller Seen on OCZ SSD

    Image representing SandForce as depicted in Cr...
    Image via CrunchBase

    Last week during CES 2011, The Tech Report spotted OCZ’s Vertex 3 Pro SSD–running in a demo system–using a next-generation SandForce SF-2582 controller and a 6Gbps Serial ATA interface. OCZ demonstrated its read and write speeds by running the ATTO Disk Benchmark which clearly showed the disk hitting sustained read speeds of 550 MB/s and sustained write speeds of 525 MB/s.

    via Next-Gen SandForce Controller Seen on OCZ SSD.

    Big news, test samples of the SandForce SF-2000 series flash memory controllers are being shown in products demoed at the Consumer Electronics Shows. And SSDs with SATA interfaces are testing through the roof. The numbers quoted for a 6GB/sec. SATA SSD are in the 500+GB/sec. range. Previously you would need to choose a PCIe based SSD drive from OCZ or Fusion-io to get anywhere near that high of  speed sustained. Combine this with the future possibility of SF-2000 being installed on future PCIe based SSDs and there’s no telling how much the throughput will scale. If four of the Vertex drives were bound together as a RAID 0 set with SF-2000 drive controllers managing it, is it possible to see a linear scaling of throughput. Could we see 2,000 MB/sec. on PCIe 8x SSD cards? And what would be the price on such a card fully configured with 1.2 TB of SSD drives? Hard to say what things may come, but just the thought of being able to buy retail versions of these makes me think a paradigm shift is in the works that neither Intel nor Microsoft are really thinking about right now.

    One comment on this article as posted on the original website, Tom’s Hardware, included the observation that the speeds quoted for this SATA 6GBps drive are approaching the memory bandwidth of several generations old PC-133 DRAM memory chips. And as I have said previously, I still have an old first generation Titanium Powerbook from Apple that uses that same memory chip standard PC-133. So given that SSD hard drives are fast approaching the speed of somewhat older main memory chips I can only say we are fast approaching a paradigm shift in desktop and enterprise computing. I dub thee, the All Solid State (ASS) era where no magnetic or rotating mechanical media enter into the equation. We run on silicon semiconductors from top to bottom, no Giant Magneto-Resistive technology necessary. Even our removable media are flash memory based USB drives we put in our pockets and walk around with on key chains.

  • CES 2011: Corsair Performance Series 3 SSD Benchmarks – AnandTech :: Your Source for Hardware Analysis and News

    Image representing SandForce as depicted in Cr...
    Image via CrunchBase

    The next wave of high end consumer SSDs will begin shipping this month, and I believe Corsair may be the first out the gate. Micron will follow shortly with its C400 and then we’ll likely see a third generation offering from Intel before eventually getting final hardware based on SandForce’s SF-2000 controllers in May.

    via CES 2011: Corsair Performance Series 3 SSD Benchmarks – AnandTech :: Your Source for Hardware Analysis and News.

    This just in from Consumer Electronics Show in Las Vegas, via Anandtech. SandForce SF-2000 scheduled to drop in May of this year. Get ready as you will see a huge upsurge in releases of new SSD products attempting to best one another in the sustained Read/Write category. And I’m not talking just SSDs but PCIe based cards with SSD RAIDs embedded on them communicating through a 2 Lane 8X PCI Express interface. I’m going to take a wild guess and say you will see products fitting this description easily hitting 700 to 900 MB/s sustained Read and Write. Prices will be on the top end of the scale as even the current shipping products all fall in to the $1200 to $1500 range. Expect the top end to be LSI based products for $15,000 or third party OEM manufacturers who might be willing to sell a fully configured 1TByte card for maybe ~$2,000. After the SF-2000 is released, I don’t know how long it will take for designers to prototype and release to manufacturing any new designs incorporating this top of the line SSD flash memory controller. It’s possible as the top end continues to increase in performance current shipping product might start to fall in price to clear out the older, lower performance designs.

  • Hitachi GST ends STEC’s monopoly • The Register

    Hitachi GST flash drives are hitting the streets and, at last, ending STEC’s monopoly in the supply of Fibre Channel interface SSDs.

    EMC startled the enterprise storage array world by embracing STEC SSDs (solid state drives) in its arrays last year as a way of dramatically lowering the latency for access to the most important data in the arrays. It has subsequently delivered FAST automated data movement across different tiers of storage in its arrays, ensuring that sysadms don’t have to involved in managing data movement at a tedious and time-consuming level.

    via Hitachi GST ends STEC’s monopoly • The Register.

    In the computer world the data center is often the measure of all things in terms of speed and performance. Time was, the disk drive interface of choice was the SCSI drive and then it’s higher speed evolutions Fast/Wide UltraSCSI. But then a new interface hit that used fibre optic cables to move storage out of the computer box to a separate box that managed all the hard drives in one spot and this was called a Storage Array. The new connector/cable combo was named Fibre Channel and it was fast, fast, fast. It become the absolute brand name off all vendors trying to sell more and more hard drives into the data center. Newer evolved versions of Fibre Channel came to market, each one slightly faster than the rest. And eventually Fibre Channel was built right into the hard drives themselves, so that you could be assured the speed was native Fibre Channel 3Gigabytes per second from one end to the other. But Fibre Channel has always been prohibitively expensive though a lot of it has been sold over the years. Volume has not brought down the price of Fibre Channel one bit in the time that it’s been the most widely deployed disk drive interface. A few competitors have cropped up the old Parallel ATA and Serial ATA drives from the desktop market have attempted to compete. And a newer SCSI drive interface called Serial Attached SCSI is now seeing some wider acceptance. However the old guard who are mentally and emotionally attached to their favorite Fibre Channel drive interface are not about to give up even has spinning disk speeds have been trumped by the almighty Flash memory based solid state drive (SSD). And a company named STEC knew it could sell a lot of SSDs if only someone could put a Fibre Channel interface on the circuit board, allaying any fears of the Fibre Channel adherents that they needed to evolve and change.

    Yes it’s true STEC was the only game in town for what I consider the Fibre Channel legacy interface in old-line Storage Array manufacturers. They have sold tons of their drives to third parties who package up their wares into turnkey ‘Enterprise’ solutions for drive arrays and cache controllers (all of which just speed up things). And being the first-est with the most-est is a good business strategy until the second source of your product comes online. So it’s always a race to sell as much as you can until the deadline hits and everyone rushes to the second source. Here now is Hitachi’s announcement they are now manufacturing an SSD with a Fibre Channel interface onboard for the Enterprise data center customers.