high_scalability high_scalability-2010 high_scalability-2010-953 knowledge-graph by maker-knowledge-mining
Source: html
Introduction: In some ways the original Amazon cloud, the one most of us still live in, was like that really cool house that when you stepped inside and saw the old green shag carpet in the living room, you knew the house hadn't been updated in a while. The network is a little slow, the processors are a bit dated, and virtualization made the house just feel smaller. It has been difficult to run high bandwidth or low latency workloads in the cloud. Bottlenecks everywhere. Not a big deal for most applications, but for many high performance applications (HPC) it was a killer. In a typical house you might just do a remodel. Upgrade a few rooms. Swap out builder quality appliances with gleaming stainless steel monsters. But Amazon has a big lot, instead of remodeling they simply keep adding on entire new wings, kind of like the Winchester Mystery House of computing. The first new wing added was a CPU based HPC system featuring blazingly fast Nehalem chips , virtualization replaced by a close t
sentIndex sentText sentNum sentScore
1 In some ways the original Amazon cloud, the one most of us still live in, was like that really cool house that when you stepped inside and saw the old green shag carpet in the living room, you knew the house hadn't been updated in a while. [sent-1, score-0.541]
2 The network is a little slow, the processors are a bit dated, and virtualization made the house just feel smaller. [sent-2, score-0.256]
3 In a typical house you might just do a remodel. [sent-6, score-0.197]
4 Most people still probably don't even know this part of the house exists. [sent-12, score-0.197]
5 The newest addition is a beauty, it's a graphics processing unit (GPU) cluster as described by Werner Vogels in Expanding the Cloud - Adding the Incredible Power of the Amazon EC2 Cluster GPU Instances . [sent-13, score-0.302]
6 To get a feeling of the speed involved read BillMcColl's comment : Cloudscale is now able to ANALYZE a SINGLE STREAM of entity events at a rate of TWO MILLION EVENTS PER SECOND (150MB/sec) on an 8-node 10-GigE Amazon cloud cluster. [sent-17, score-0.142]
7 To be able build applications that exploit this level of parallelism one needs to enter a very specific mindset of kernels, kernel functions, threads-blocks, grids of threads-blocks, mapping to hierarchical memory, etc. [sent-23, score-0.342]
8 There are a number of techniques that every programmer has grown up with, such as branching, that are not available, or should be avoided on GPUs if one wants to truly exploit its power. [sent-25, score-0.185]
9 Modern CPUs strongly favor lower latency of operations with clock cycles in the nanoseconds and we have built general purpose software architectures that can exploit these low latencies very well. [sent-26, score-0.486]
10 Now that our ability to generate higher and higher clock rates has stalled and CPU architectural improvements have shifted focus towards multiple cores, we see that it is becoming harder to efficiently use these computer systems. [sent-27, score-0.2]
11 One trade-off area where our general purpose CPUs were not performing well was that of massive fine grain parallelism. [sent-28, score-0.214]
12 The throughput of this pipeline is more important than the latency of the individual operations . [sent-30, score-0.136]
13 Because of its focus on latency, the generic CPU yielded rather inefficient system for graphics processing. [sent-31, score-0.171]
14 This lead to the birth of the Graphics Processing Unit (GPU) which was focused on providing a very fine grained parallel model, with processing organized in multiple stages, where the data would flow through. [sent-32, score-0.127]
15 The model of a GPU is that of task parallelism describing the different stages in the pipeline, as well as data parallelism within each stage, resulting in a highly efficient, high throughput computation architecture. [sent-33, score-0.401]
16 The ACM has a timely article about using GPUs for high performance computing ACM: Understanding Throughput-Oriented Architecture by Michael Garland and David Kirk: Scalability is the programmer's central concern in designing efficient algorithms for throughput-oriented machines. [sent-34, score-0.132]
17 Today's architectural trends clearly favor increasing parallelism, and effective algorithmic techniques must scale with hardware parallelism. [sent-35, score-0.207]
18 Some techniques suitable for four parallel threads may be entirely unsuitable for 4,000 parallel threads. [sent-36, score-0.137]
19 Second, for a surprisingly large number of problems there is now a ready supply of GPU supercomputeryness. [sent-42, score-0.076]
20 But that's kind of boring :-) Clearly GPUs aren't general purpose processors . [sent-52, score-0.273]
wordName wordTfidf (topN-words)
[('gpu', 0.471), ('gpus', 0.348), ('house', 0.197), ('graphics', 0.171), ('parallelism', 0.161), ('shag', 0.147), ('hpc', 0.133), ('exploit', 0.115), ('acm', 0.098), ('purpose', 0.089), ('vogels', 0.083), ('stages', 0.079), ('favor', 0.078), ('clock', 0.078), ('cpu', 0.077), ('supply', 0.076), ('cloud', 0.076), ('pipeline', 0.076), ('techniques', 0.07), ('algorithms', 0.068), ('processing', 0.068), ('dated', 0.067), ('cloudscale', 0.067), ('garland', 0.067), ('eventing', 0.067), ('gpgpus', 0.067), ('unsuitable', 0.067), ('events', 0.066), ('kernel', 0.066), ('general', 0.066), ('efficient', 0.064), ('unit', 0.063), ('kirk', 0.063), ('blazingly', 0.063), ('stalled', 0.063), ('steel', 0.063), ('articlesa', 0.063), ('retrospect', 0.063), ('oriented', 0.062), ('latency', 0.06), ('exotic', 0.06), ('manycore', 0.06), ('kind', 0.059), ('cpus', 0.059), ('architectural', 0.059), ('fine', 0.059), ('processors', 0.059), ('builder', 0.057), ('nails', 0.057), ('sciences', 0.057)]
simIndex simValue blogId blogTitle
same-blog 1 1.0000001 953 high scalability-2010-12-03-GPU vs CPU Smackdown : The Rise of Throughput-Oriented Architectures
Introduction: In some ways the original Amazon cloud, the one most of us still live in, was like that really cool house that when you stepped inside and saw the old green shag carpet in the living room, you knew the house hadn't been updated in a while. The network is a little slow, the processors are a bit dated, and virtualization made the house just feel smaller. It has been difficult to run high bandwidth or low latency workloads in the cloud. Bottlenecks everywhere. Not a big deal for most applications, but for many high performance applications (HPC) it was a killer. In a typical house you might just do a remodel. Upgrade a few rooms. Swap out builder quality appliances with gleaming stainless steel monsters. But Amazon has a big lot, instead of remodeling they simply keep adding on entire new wings, kind of like the Winchester Mystery House of computing. The first new wing added was a CPU based HPC system featuring blazingly fast Nehalem chips , virtualization replaced by a close t
Introduction: All in all this is still my favorite post and I still think it's an accurate vision of a future. Not everyone agrees, but I guess we'll see..."But it is not complicated. [There's] just a lot of it." \--Richard Feynmanon how the immense variety of the world arises from simple rules.Contents:Have We Reached the End of Scaling?Applications Become Black Boxes Using Markets to Scale and Control CostsLet's Welcome our Neo-Feudal OverlordsThe Economic Argument for the Ambient CloudWhat Will Kill the Cloud?The Amazing Collective Compute Power of the Ambient CloudUsing the Ambient Cloud as an Application RuntimeApplications as Virtual StatesConclusionWe have not yet begun to scale. The world is still fundamentally disconnected and for all our wisdom we are still in the earliest days of learning how to build truly large planet-scaling applications.Today 350 million users on Facebook is a lot of users and five million followers on Twitter is a lot of followers. This may seem like a lot now, but c
Introduction: "But it is not complicated. [There's] just a lot of it." \--Richard Feynmanon how the immense variety of the world arises from simple rules.Contents:Have We Reached the End of Scaling?Applications Become Black Boxes Using Markets to Scale and Control CostsLet's Welcome our Neo-Feudal OverlordsThe Economic Argument for the Ambient CloudWhat Will Kill the Cloud?The Amazing Collective Compute Power of the Ambient CloudUsing the Ambient Cloud as an Application RuntimeApplications as Virtual StatesConclusionWe have not yet begun to scale. The world is still fundamentally disconnected and for all our wisdom we are still in the earliest days of learning how to build truly large planet-scaling applications.Today 350 million users on Facebook is a lot of users and five million followers on Twitter is a lot of followers. This may seem like a lot now, but consider we have no planet wide applications yet. None.Tomorrow the numbers foreshadow a newCambrian explosionof connectivity that will look as
4 0.15603702 636 high scalability-2009-06-23-Learn How to Exploit Multiple Cores for Better Performance and Scalability
Introduction: InfoQueue has this excellent talk by Brian Goetz on the new features being added to Java SE 7 that will allow programmers to fully exploit our massively multi-processor future. While the talk is about Java it's really more general than that and there's a lot to learn here for everyone. Brian starts with a short, coherent, and compelling explanation of why programmers can't expect to be saved by ever faster CPUs and why we must learn to exploit the strengths of multiple core computers to make our software go faster. Some techniques for exploiting multiple cores are given in an equally short, coherent, and compelling explanation of why divide and conquer as the secret to multi-core bliss, fork-join, how the Java approach differs from map-reduce, and lots of other juicy topics. The multi-core "problem" is only going to get worse. Tilera founder Anant Agarwal estimates by 2017 embedded processors could have 4,096 cores, server CPUs might have 512 cores and desktop chips could use
5 0.1502195 661 high scalability-2009-07-25-Latency is Everywhere and it Costs You Sales - How to Crush it
Introduction: Update 8 : The Cost of Latency by James Hamilton. James summarizing some latency info from Steve Souder , Greg Linden , and Marissa Mayer . Speed [is] an undervalued and under-discussed asset on the web. Update 7: How do you know when you need more memcache servers? . Dathan Pattishall talks about using memcache not to scale, but to reduce latency and reduce I/O spikes, and how to use stats to know when more servers are needed. Update 6: Stock Traders Find Speed Pays, in Milliseconds . Goldman Sachs is making record profits off a 500 millisecond trading advantage. Yes, latency matters. As an interesting aside, Libet found 500 msecs is about the time it takes the brain to weave together an experience of consciousness from all our sensor inputs. Update 5: Shopzilla's Site Redo - You Get What You Measure . At the Velocity conference Phil Dixon, from Shopzilla, presented data showing a 5 second speed up resulted in a 25% increase in page views, a 10% increas
6 0.1477105 609 high scalability-2009-05-28-Scaling PostgreSQL using CUDA
7 0.1461542 538 high scalability-2009-03-16-Are Cloud Based Memory Architectures the Next Big Thing?
8 0.13813703 1621 high scalability-2014-03-28-Stuff The Internet Says On Scalability For March 28th, 2014
10 0.13132547 1024 high scalability-2011-04-15-Stuff The Internet Says On Scalability For April 15, 2011
11 0.12875395 1036 high scalability-2011-05-06-Stuff The Internet Says On Scalability For May 6th, 2011
12 0.12328783 1240 high scalability-2012-05-07-Startups are Creating a New System of the World for IT
13 0.12249038 608 high scalability-2009-05-27-The Future of the Parallelism and its Challenges
14 0.12059554 1141 high scalability-2011-11-11-Stuff The Internet Says On Scalability For November 11, 2011
15 0.11646696 768 high scalability-2010-02-01-What Will Kill the Cloud?
16 0.11622469 1186 high scalability-2012-02-02-The Data-Scope Project - 6PB storage, 500GBytes-sec sequential IO, 20M IOPS, 130TFlops
17 0.11420655 413 high scalability-2008-10-14-Sun Storage and Archive Solution for HPC
18 0.11371709 1487 high scalability-2013-07-05-Stuff The Internet Says On Scalability For July 5, 2013
19 0.11086286 778 high scalability-2010-02-15-The Amazing Collective Compute Power of the Ambient Cloud
20 0.11007769 1456 high scalability-2013-05-13-The Secret to 10 Million Concurrent Connections -The Kernel is the Problem, Not the Solution
topicId topicWeight
[(0, 0.204), (1, 0.092), (2, 0.039), (3, 0.101), (4, -0.073), (5, 0.005), (6, 0.041), (7, 0.04), (8, -0.065), (9, 0.027), (10, 0.009), (11, -0.005), (12, 0.03), (13, 0.028), (14, -0.007), (15, -0.018), (16, 0.003), (17, 0.004), (18, 0.01), (19, 0.034), (20, 0.011), (21, -0.018), (22, -0.036), (23, 0.006), (24, 0.04), (25, -0.005), (26, 0.008), (27, 0.019), (28, 0.059), (29, 0.003), (30, 0.013), (31, 0.063), (32, -0.007), (33, 0.013), (34, 0.024), (35, -0.029), (36, 0.061), (37, 0.009), (38, 0.012), (39, 0.002), (40, -0.067), (41, 0.036), (42, -0.058), (43, -0.01), (44, 0.045), (45, 0.001), (46, 0.001), (47, -0.058), (48, 0.011), (49, 0.018)]
simIndex simValue blogId blogTitle
same-blog 1 0.95965433 953 high scalability-2010-12-03-GPU vs CPU Smackdown : The Rise of Throughput-Oriented Architectures
Introduction: In some ways the original Amazon cloud, the one most of us still live in, was like that really cool house that when you stepped inside and saw the old green shag carpet in the living room, you knew the house hadn't been updated in a while. The network is a little slow, the processors are a bit dated, and virtualization made the house just feel smaller. It has been difficult to run high bandwidth or low latency workloads in the cloud. Bottlenecks everywhere. Not a big deal for most applications, but for many high performance applications (HPC) it was a killer. In a typical house you might just do a remodel. Upgrade a few rooms. Swap out builder quality appliances with gleaming stainless steel monsters. But Amazon has a big lot, instead of remodeling they simply keep adding on entire new wings, kind of like the Winchester Mystery House of computing. The first new wing added was a CPU based HPC system featuring blazingly fast Nehalem chips , virtualization replaced by a close t
2 0.83819258 636 high scalability-2009-06-23-Learn How to Exploit Multiple Cores for Better Performance and Scalability
Introduction: InfoQueue has this excellent talk by Brian Goetz on the new features being added to Java SE 7 that will allow programmers to fully exploit our massively multi-processor future. While the talk is about Java it's really more general than that and there's a lot to learn here for everyone. Brian starts with a short, coherent, and compelling explanation of why programmers can't expect to be saved by ever faster CPUs and why we must learn to exploit the strengths of multiple core computers to make our software go faster. Some techniques for exploiting multiple cores are given in an equally short, coherent, and compelling explanation of why divide and conquer as the secret to multi-core bliss, fork-join, how the Java approach differs from map-reduce, and lots of other juicy topics. The multi-core "problem" is only going to get worse. Tilera founder Anant Agarwal estimates by 2017 embedded processors could have 4,096 cores, server CPUs might have 512 cores and desktop chips could use
3 0.7936765 826 high scalability-2010-05-12-The Rise of the Virtual Cellular Machines
Introduction: My apologies if you were looking for a post about cell phones. This post is about high density nanodevices. It's a follow up to How will memristors change everything? for those wishing to pursue these revolutionary ideas in more depth. This is one of those areas where if you are in the space then there's a lot of available information and if you are on the outside then it doesn't even seem to exist. Fortunately, Ben Chandler from The SyNAPSE Project , was kind enough to point me to a great set of presentations given at the 12th IEEE CNNA - International Workshop on Cellular Nanoscale Networks and their Applications - Towards Megaprocessor Computing. WARNING: these papers contain extreme technical content. If you are like me and you aren't an electrical engineer, much of it may make a sort of surface sense, but the deep and twisty details will fly over head. For the more software minded there are a couple more accessible presentations: Intelligent Machines built with Memristiv
4 0.77522063 1127 high scalability-2011-09-28-Pursue robust indefinite scalability with the Movable Feast Machine
Introduction: And now for something completely different, brought to you by David Ackley and Daniel Cannon in their playfully thought provoking paper: Pursue robust indefinite scalability , wherein they try to take a fresh look at neural networks, starting from scratch. What is this strange thing called indefinite scalability ? They sound like words that don't really go together: Indefinite scalability is the property that the design can support open-ended computational growth without substantial re-engineering, in as strict as sense as can be managed. By comparison, many computer, algorithm, and network designs -- even those that address scalability -- are only finitely scalable because their scalability occurs within some finite space. For example, an in-core sorting algorithm for a 32 bit machine can only scale to billions of numbers before address space is exhausted and then that algorithm must be re-engineered. Our idea is to expose indefinitely scalable computational power to program
5 0.77193785 1600 high scalability-2014-02-21-Stuff The Internet Says On Scalability For February 21st, 2014
Introduction: Hey, it's HighScalability time (a particularly bountiful week): The Telephone Wires of Manhattan in 1887 ( full ) $19 billion: you know what it is; $46 billion : cost of Sochi Olympics; 400 gigabytes : data transmitted during the Sochi opening ceremony; 26.9 million : Stack Overflow community monthly visitors; 93 million : Candy Crush daily active users; 200-400 Gbps : The New Normal in DDoS Attacks Quotable Quotes: @brianacton : Facebook turned me down. It was a great opportunity to connect with some fantastic people. Looking forward to life's next adventure. @BenedictEvans : Flickr: $35m. Youtube: $1.65bn Whatsapp: $19bn. Mobile is big. And global. And the next computing platform. Paying attention? @taziden : On the Internet, worst cases will become common cases #fosdem #postfix Brian Hayes : Any quantum program must have a stovepipe architecture: Information flows straight through. So you think V
6 0.76528215 581 high scalability-2009-04-26-Map-Reduce for Machine Learning on Multicore
7 0.76441753 608 high scalability-2009-05-27-The Future of the Parallelism and its Challenges
8 0.74620366 1572 high scalability-2014-01-03-Stuff The Internet Says On Scalability For January 3rd, 2014
9 0.74235094 317 high scalability-2008-05-10-Hitting 300 SimbleDB Requests Per Second on a Small EC2 Instance
10 0.73780686 1436 high scalability-2013-04-05-Stuff The Internet Says On Scalability For April 5, 2013
11 0.73436564 612 high scalability-2009-05-31-Parallel Programming for real-world
12 0.73068613 1204 high scalability-2012-03-06-Ask For Forgiveness Programming - Or How We'll Program 1000 Cores
13 0.72735131 823 high scalability-2010-05-05-How will memristors change everything?
14 0.72590452 1218 high scalability-2012-03-29-Strategy: Exploit Processor Affinity for High and Predictable Performance
15 0.72501105 534 high scalability-2009-03-12-Google TechTalk: Amdahl's Law in the Multicore Era
16 0.72234839 1581 high scalability-2014-01-17-Stuff The Internet Says On Scalability For January 17th, 2014
17 0.72209138 660 high scalability-2009-07-21-Paper: Parallelizing the Web Browser
18 0.72025758 1235 high scalability-2012-04-27-Stuff The Internet Says On Scalability For April 27, 2012
19 0.71965462 604 high scalability-2009-05-20-Paper: Flux: An Adaptive Partitioning Operator for Continuous Query Systems
20 0.71549201 1177 high scalability-2012-01-19-Is it time to get rid of the Linux OS model in the cloud?
topicId topicWeight
[(1, 0.152), (2, 0.206), (10, 0.056), (40, 0.025), (51, 0.173), (52, 0.011), (56, 0.02), (61, 0.04), (79, 0.12), (85, 0.042), (94, 0.074)]
simIndex simValue blogId blogTitle
1 0.97894019 1168 high scalability-2012-01-04-How Facebook Handled the New Year's Eve Onslaught
Introduction: How does Facebook handle the massive New Year's Eve traffic spike? Thanks to Mike Swift, in Facebook gets ready for New Year's Eve , we get a little insight as to their method for the madness, nothing really detailed, but still interesting. Problem Setup Facebook expects tha one billion+ photos will be shared on New Year's eve. Facebook's 800 million users are scattered around the world. Three quarters live outside the US. Each user is linked to an average of 130 friends. Photos and posts must appear in less than a second. Opening a homepage requires executing requests on a 100 different servers, and those requests have to be ranked, sorted, and privacy-checked, and then rendered. Different events put different stresses on different parts of Facebook. Photo and Video Uploads - Holidays require hundreds of terabytes of capacity News Feed - News events like big sports events and the death of Steve Jobs drive user status updates Coping Strategies Try
same-blog 2 0.93702853 953 high scalability-2010-12-03-GPU vs CPU Smackdown : The Rise of Throughput-Oriented Architectures
Introduction: In some ways the original Amazon cloud, the one most of us still live in, was like that really cool house that when you stepped inside and saw the old green shag carpet in the living room, you knew the house hadn't been updated in a while. The network is a little slow, the processors are a bit dated, and virtualization made the house just feel smaller. It has been difficult to run high bandwidth or low latency workloads in the cloud. Bottlenecks everywhere. Not a big deal for most applications, but for many high performance applications (HPC) it was a killer. In a typical house you might just do a remodel. Upgrade a few rooms. Swap out builder quality appliances with gleaming stainless steel monsters. But Amazon has a big lot, instead of remodeling they simply keep adding on entire new wings, kind of like the Winchester Mystery House of computing. The first new wing added was a CPU based HPC system featuring blazingly fast Nehalem chips , virtualization replaced by a close t
Introduction: We are not yet at the End of History for database theory as Peter Bailis and the Database Group at UC Berkeley continue to prove with a great companion blog post to their new paper: Scalable Atomic Visibility with RAMP Transactions . I like the approach of pairing a blog post with a paper. A paper almost by definition is formal, but a blog post can help put a paper in context and give it some heart. From the abstract: Databases can provide scalability by partitioning data across several servers. However, multi-partition, multi-operation transactional access is often expensive, employing coordination-intensive locking, validation, or scheduling mechanisms. Accordingly, many real-world systems avoid mechanisms that provide useful semantics for multi-partition operations . This leads to incorrect behavior for a large class of applications including secondary indexing, foreign key enforcement, and materialized view maintenance . In this work, we identify a new isolation mode
4 0.9331516 741 high scalability-2009-11-16-Building Scalable Systems Using Data as a Composite Material
Introduction: Think of building websites as engineering composite materials . A composite material is when two or more materials are combined to create a third material that does something useful that the components couldn't do on their own. Composites like reinforced concrete have revolutionized design and construction. When building websites we usually bring different component materials together, like creating a composite, to get the features we need rather than building a completely new thing from scratch that does everything we want. This approach has been seen as a hack because it leads to inelegancies like data duplication; great gobs of component glue; consistency issues; and messy operations. But what if the the composite approach is really a strength, not a hack, but a messy part of the world that needs to be embraced rather than belittled? They key is to see data as a material . Right now we are arguing which is the best single material to build with. Is it NoSQL , relational, mass
5 0.93215877 1134 high scalability-2011-10-28-Stuff The Internet Says On Scalability For October 28, 2011
Introduction: You deserve a HighScalability today : S3: 566 Billion Objects, 370K requests/sec ; Titan: 38,400-processor, 20-petaflop 1,000,000 daily users and no cache . Wooga flash game with 50K DB updates/second, Ruby backend. They hit an IO wall with MySQL at 1000 DB updates/sec. They needed more so they went with Redis. Not quite honest to say no cache was used as everything is RAM, but maybe that's the point. Use a lot of automation. Inactive users are archived. Moved away from EBS. Making dynamic sitesscale like static sites by Wim Godden. Use Varnish, Nginx, and memcached. The Lifecycle of a Web Page on StumbleUpon infographic. 2.2 mllion web pages are added to StumbleUpon each month. Nice discussion of bounce rate . James Hamilton with an excellent overview of the Storage Infrastructure Behind Facebook Messages . That's 6B+ messages a day. Scaling Twilio . Twilio has scaled traffic by more 100x over the past year, and expanded their server infrastructure from a
6 0.92932504 98 high scalability-2007-09-18-Sync data on all servers
8 0.90624654 481 high scalability-2009-01-02-Strategy: Understanding Your Data Leads to the Best Scalability Solutions
9 0.90062213 138 high scalability-2007-10-30-Feedblendr Architecture - Using EC2 to Scale
10 0.89583492 1644 high scalability-2014-05-07-Update on Disqus: It's Still About Realtime, But Go Demolishes Python
11 0.88136452 1638 high scalability-2014-04-28-How Disqus Went Realtime with 165K Messages Per Second and Less than .2 Seconds Latency
12 0.87741745 63 high scalability-2007-08-09-Lots of questions for high scalability - high availability
13 0.86320144 327 high scalability-2008-05-27-How I Learned to Stop Worrying and Love Using a Lot of Disk Space to Scale
15 0.85807222 357 high scalability-2008-07-26-Google's Paxos Made Live – An Engineering Perspective
16 0.85755467 1186 high scalability-2012-02-02-The Data-Scope Project - 6PB storage, 500GBytes-sec sequential IO, 20M IOPS, 130TFlops
17 0.85576546 1027 high scalability-2011-04-20-Packet Pushers: How to Build a Low Cost Data Center
18 0.85520434 1581 high scalability-2014-01-17-Stuff The Internet Says On Scalability For January 17th, 2014
19 0.85320079 789 high scalability-2010-03-05-Strategy: Planning for a Power Outage Google Style
20 0.85312653 1122 high scalability-2011-09-23-Stuff The Internet Says On Scalability For September 23, 2011