high_scalability high_scalability-2014 high_scalability-2014-1594 knowledge-graph by maker-knowledge-mining
Source: html
Introduction: In the scalability is specialization department here is an interesting paper presented at HotNets '13 on high performance networking: Network Stack Specialization for Performance . The idea is generalizing a service so it fits in the kernel comes at a high performance cost. So move TCP into user space. The result is a web server with ~3.5x the throughput of Nginx "while experiencing low CPU utilization, linear scaling on multicore systems, and saturating current NIC hardware." Here's a good description of the paper published on Layer 9 : Traditionally, servers and OSes have been built to be general purpose. However now we have a high degree of specialization. In fact, in a big web service, you might have thousands of machines dedicated to one function. Therefore, there's scope for specialization. This paper looks at a specific opportunity in that space. Network stacks today are good for high throughput with large transfers, but not small files (which are common in web browsi
sentIndex sentText sentNum sentScore
1 In the scalability is specialization department here is an interesting paper presented at HotNets '13 on high performance networking: Network Stack Specialization for Performance . [sent-1, score-0.363]
2 The idea is generalizing a service so it fits in the kernel comes at a high performance cost. [sent-2, score-0.255]
3 5x the throughput of Nginx "while experiencing low CPU utilization, linear scaling on multicore systems, and saturating current NIC hardware. [sent-5, score-0.239]
4 " Here's a good description of the paper published on Layer 9 : Traditionally, servers and OSes have been built to be general purpose. [sent-6, score-0.208]
5 However now we have a high degree of specialization. [sent-7, score-0.078]
6 This paper looks at a specific opportunity in that space. [sent-10, score-0.208]
7 Network stacks today are good for high throughput with large transfers, but not small files (which are common in web browsing). [sent-11, score-0.488]
8 For example one of their experiments showed with 8KB files, ~85% CPU load for just ~50% throughput (5 Gbps) . [sent-12, score-0.139]
9 The goal of this paper is to design a specialized network stack that addresses that problem, by reducing system overhead and providing free flow of memory between the NIC and applications. [sent-13, score-0.926]
10 The paper introduces a zero copy web server called Sandstorm which accomplishes that. [sent-14, score-0.304]
11 Key design decisions include putting apps and the network stack in the same address space; static content pre-segmented into packets; batching of data; and bufferless operation. [sent-15, score-0.382]
12 In performance evaluation, the most impressive improvements are in flow completion time when there are a small number of connections, and CPU load reduction of about 5x . [sent-16, score-0.357]
13 When moving to 6x10GbE NICs, the thoughput improvement also becomes impressive: about 3. [sent-17, score-0.113]
14 5x higher throughput than FreeBSD+nginx and even bigger improvement over Linux+nginx. [sent-18, score-0.252]
15 Overall, the paper's contribution is a new programming model that tightly couples the application with the network stack, reaching 55 Gbps at 72% load . [sent-19, score-0.388]
16 General-purpose operating system stacks have been around a long time, and have demonstrated the ability to transcend multiple generations of hardware. [sent-22, score-0.595]
17 We examined performance on servers manufactured seven years apart, and demonstrated that although the performance bottlenecks were now in different places, the same design delivered significant benefits on both platforms. [sent-24, score-0.615]
18 It is our belief that a blend of specialized network stacks with performance-critical applications such as web services, can dramatically improve the scalability of current hardware, reducing costs and energy demands, offering a practical alternative to general-purpose designs. [sent-25, score-0.823]
wordName wordTfidf (topN-words)
[('stacks', 0.271), ('sandstorm', 0.246), ('paper', 0.208), ('specialization', 0.173), ('gbps', 0.156), ('specialized', 0.153), ('nic', 0.148), ('demonstrated', 0.147), ('stack', 0.145), ('throughput', 0.139), ('network', 0.126), ('impressive', 0.12), ('improvement', 0.113), ('specializations', 0.112), ('articlesis', 0.112), ('manufactured', 0.112), ('publishedon', 0.112), ('design', 0.111), ('nginx', 0.105), ('transcend', 0.1), ('promoting', 0.1), ('generalizing', 0.1), ('saturating', 0.1), ('tcp', 0.097), ('reducing', 0.097), ('accomplishes', 0.096), ('ow', 0.096), ('couples', 0.093), ('signi', 0.093), ('bene', 0.093), ('blend', 0.091), ('cant', 0.091), ('nics', 0.091), ('contribution', 0.091), ('ts', 0.091), ('examined', 0.091), ('freebsd', 0.089), ('flow', 0.086), ('browsing', 0.085), ('belief', 0.085), ('ll', 0.082), ('webserver', 0.082), ('noticeable', 0.08), ('connections', 0.078), ('tightly', 0.078), ('high', 0.078), ('generations', 0.077), ('performance', 0.077), ('pipe', 0.074), ('completion', 0.074)]
simIndex simValue blogId blogTitle
same-blog 1 0.99999976 1594 high scalability-2014-02-12-Paper: Network Stack Specialization for Performance
Introduction: In the scalability is specialization department here is an interesting paper presented at HotNets '13 on high performance networking: Network Stack Specialization for Performance . The idea is generalizing a service so it fits in the kernel comes at a high performance cost. So move TCP into user space. The result is a web server with ~3.5x the throughput of Nginx "while experiencing low CPU utilization, linear scaling on multicore systems, and saturating current NIC hardware." Here's a good description of the paper published on Layer 9 : Traditionally, servers and OSes have been built to be general purpose. However now we have a high degree of specialization. In fact, in a big web service, you might have thousands of machines dedicated to one function. Therefore, there's scope for specialization. This paper looks at a specific opportunity in that space. Network stacks today are good for high throughput with large transfers, but not small files (which are common in web browsi
2 0.15225105 1177 high scalability-2012-01-19-Is it time to get rid of the Linux OS model in the cloud?
Introduction: You program in a dynamic language, that runs on a JVM, that runs on a OS designed 40 years ago for a completely different purpose, that runs on virtualized hardware. Does this make sense? We've talked about this idea before in Machine VM + Cloud API - Rewriting The Cloud From Scratch , where the vision is to treat cloud virtual hardware as a compiler target, and converting high-level language source code directly into kernels that run on it. As new technologies evolve the friction created by our old tool chains and architecture models becomes ever more obvious. Take, for example, what a team at UCSD is releasing: a phase-change memory prototype - a solid state storage device that provides performance thousands of times faster than a conventional hard drive and up to seven times faster than current state-of-the-art solid-state drives (SSDs). However, PCM has access latencies several times slower than DRAM. This technology has obvious mind blowing implications, but an
3 0.12061736 314 high scalability-2008-05-03-Product: nginx
Introduction: Update 6 : nginx_http_push_module . Turn nginx into a long-polling message queuing HTTP push server. Update 5 : In Load Balancer Update Barry describes how WordPress.com moved from Pound to Nginx and are now "regularly serving about 8-9k requests/second and about 1.2Gbit/sec through a few Nginx instances and have plenty of room to grow!". Update 4 : Nginx better than Pound for load balancing. Pound spikes at 80% CPU, Nginx uses 3% and is easier to understand and better documented. Update 3 : igvita.com combines two cool tools together for better performance in Nginx and Memcached, a 400% boost! . Update 2 : Software Project on Installing Nginx Web Server w/ PHP and SSL . Breaking away from mother Apache can be a scary proposition and this kind of getting started article really helps easy the separation. Update: Slicehost has some nice tutorials on setting up Nginx . From their website: Nginx ("engine x") is a high-performance HTTP server and reverse proxy, as wel
4 0.11457416 1596 high scalability-2014-02-14-Stuff The Internet Says On Scalability For February 14th, 2014
Introduction: Hey, it's HighScalability time: Climbing the World's Second Tallest Building 5 billion : Number of phone records NSA collects per day; Facebook : 1.23 billion users, 201.6 billion friend connections, 400 billion shared photos, and 7.8 trillion messages sent since the start of 2012. Quotable Quotes: @ShrikanthSS : people repeatedly underestimate the cost of busy waits @mcclure111 : Learning today java․net․URL․equals is a blocking operation that hits the network shook me badly. I don't know if I can trust the world now. @hui_kenneth : @randybias: “3 ways 2 be market leader - be 1st, be best, or be cheapest. #AWS was all 3. Now #googlecloud may be best & is the cheapest.” @thijs : The nice thing about Paper is that we can point out to clients that it took 18 experienced designers and developers two years to build. @neil_conway : My guess is that the split between Spanner and F1 is a great example of Conway's Law. How F
5 0.11303419 784 high scalability-2010-02-25-Paper: High Performance Scalable Data Stores
Introduction: The world of scalable databases is not a simple one. They come in every race, creed, and color. Rick Cattell has brought some harmony to that world by publishing High Performance Scalable Data Stores , a nicely detailed one stop shop paper comparing scalable databases soley on the content of their character. Ironically, the first step in that evaluation is dividing the world into four groups: Key-value stores: Redis, Scalaris, Voldmort, and Riak. Document stores: Couch DB, MongoDB, and SimpleDB. Record stores: BigTable, HBase, HyperTable, and Cassandra. Scalable RDBMSs: MySQL Cluster, ScaleDB, Drizzle, and VoltDB. The paper describes each system and then compares them on the dimensions of Concurrency Control, Data Storage Replication, Transaction Model, General Comments, Maturity, K-hits, License Language. And the winner is: there are no winners. Yet. Rick concludes by pointing to a great convergence: I believe that a few of these systems will gain critical mass an
7 0.10573212 1520 high scalability-2013-09-20-Stuff The Internet Says On Scalability For September 20, 2013
8 0.10465468 1240 high scalability-2012-05-07-Startups are Creating a New System of the World for IT
9 0.10228164 661 high scalability-2009-07-25-Latency is Everywhere and it Costs You Sales - How to Crush it
10 0.10189456 1255 high scalability-2012-06-01-Stuff The Internet Says On Scalability For June 1, 2012
11 0.10110667 1316 high scalability-2012-09-04-Changing Architectures: New Datacenter Networks Will Set Your Code and Data Free
12 0.1005239 1319 high scalability-2012-09-10-Russ’ 10 Ingredient Recipe for Making 1 Million TPS on $5K Hardware
13 0.098702528 572 high scalability-2009-04-16-Paper: The End of an Architectural Era (It’s Time for a Complete Rewrite)
16 0.094571881 671 high scalability-2009-08-05-Stack Overflow Architecture
20 0.090150803 1129 high scalability-2011-09-30-Stuff The Internet Says On Scalability For September 30, 2011
topicId topicWeight
[(0, 0.191), (1, 0.059), (2, 0.005), (3, 0.023), (4, -0.054), (5, -0.001), (6, 0.047), (7, 0.027), (8, -0.098), (9, 0.064), (10, 0.005), (11, -0.057), (12, -0.0), (13, -0.017), (14, -0.027), (15, 0.02), (16, 0.074), (17, 0.031), (18, -0.009), (19, -0.061), (20, 0.045), (21, 0.048), (22, -0.065), (23, -0.029), (24, 0.045), (25, 0.039), (26, -0.063), (27, -0.065), (28, -0.005), (29, 0.001), (30, 0.008), (31, 0.026), (32, -0.018), (33, 0.04), (34, 0.017), (35, 0.039), (36, 0.01), (37, 0.037), (38, -0.018), (39, 0.04), (40, -0.029), (41, -0.002), (42, -0.033), (43, -0.065), (44, 0.002), (45, 0.009), (46, -0.087), (47, -0.058), (48, 0.006), (49, -0.032)]
simIndex simValue blogId blogTitle
same-blog 1 0.95978367 1594 high scalability-2014-02-12-Paper: Network Stack Specialization for Performance
Introduction: In the scalability is specialization department here is an interesting paper presented at HotNets '13 on high performance networking: Network Stack Specialization for Performance . The idea is generalizing a service so it fits in the kernel comes at a high performance cost. So move TCP into user space. The result is a web server with ~3.5x the throughput of Nginx "while experiencing low CPU utilization, linear scaling on multicore systems, and saturating current NIC hardware." Here's a good description of the paper published on Layer 9 : Traditionally, servers and OSes have been built to be general purpose. However now we have a high degree of specialization. In fact, in a big web service, you might have thousands of machines dedicated to one function. Therefore, there's scope for specialization. This paper looks at a specific opportunity in that space. Network stacks today are good for high throughput with large transfers, but not small files (which are common in web browsi
2 0.77798367 1213 high scalability-2012-03-22-Paper: Revisiting Network I-O APIs: The netmap Framework
Introduction: Here's a really good article in the Communications of the ACM on reducing network packet processing overhead by redesigning the network stack: Revisiting Network I/O APIs: The Netmap Framework by Luigi Rizzo . As commodity networking performance increases operating systems need to keep up or all those CPUs will go to waste. How do they make this happen? Abstract: Today 10-gigabit interfaces are used more and more in datacenters and servers. On these links, packets flow as fast as one every 67.2 nanoseconds, yet modern operating systems can take 10-20 times longer just to move one packet between the wire and the application. We can do much better, not with more powerful hardware but by revising architectural decisions made long ago regarding the design of device drivers and network stacks. The netmap framework is a promising step in this direction. Thanks to a careful design and the engineering of a new packet I/O API, netmap eliminates much unnecessary overhead and moves
3 0.76204878 1478 high scalability-2013-06-19-Paper: MegaPipe: A New Programming Interface for Scalable Network I-O
Introduction: The paper MegaPipe: A New Programming Interface for Scalable Network I/O ( video , slides ) hits the common theme that if you want to go faster you need a better car design, not just a better driver. So that's why the authors started with a clean-slate and designed a network API from the ground up with support for concurrent I/O, a requirement for achieving high performance while scaling to large numbers of connections per thread, multiple cores, etc. What they created is MegaPipe, "a new network programming API for message-oriented workloads to avoid the performance issues of BSD Socket API." The result: MegaPipe outperforms baseline Linux between 29% (for long connections) and 582% (for short connections) . MegaPipe improves the performance of a modified version of memcached between 15% and 320% . For a workload based on real-world HTTP traces, MegaPipe boosts the throughput of nginx by 75% . What's this most excellent and interesting paper about? Message-oriented netwo
4 0.74100906 1177 high scalability-2012-01-19-Is it time to get rid of the Linux OS model in the cloud?
Introduction: You program in a dynamic language, that runs on a JVM, that runs on a OS designed 40 years ago for a completely different purpose, that runs on virtualized hardware. Does this make sense? We've talked about this idea before in Machine VM + Cloud API - Rewriting The Cloud From Scratch , where the vision is to treat cloud virtual hardware as a compiler target, and converting high-level language source code directly into kernels that run on it. As new technologies evolve the friction created by our old tool chains and architecture models becomes ever more obvious. Take, for example, what a team at UCSD is releasing: a phase-change memory prototype - a solid state storage device that provides performance thousands of times faster than a conventional hard drive and up to seven times faster than current state-of-the-art solid-state drives (SSDs). However, PCM has access latencies several times slower than DRAM. This technology has obvious mind blowing implications, but an
5 0.74070543 1419 high scalability-2013-03-07-It's a VM Wasteland - A Near Optimal Packing of VMs to Machines Reduces TCO by 22%
Introduction: In Algorithm Design for Performance Aware VM Consolidation we learn some shocking facts (gambling in Casablanca?): Average server utilization in many data centers is low, estimated between 5% and 15%. This is wasteful because an idle server often consumes more than 50% of peak power. Surely that's just for old style datacenters? Nope. In Google data centers, workloads that are consolidated use only 50% of the processor cores. Every other processor core is left unused simply to ensure that performance does not degrade. It's a VM wasteland. The goal is to reduce waste by packing VMs onto machines without hurting performance or wasting resources. The idea is to select VMs that interfere the least with each other and places them together on the same server. It's a NP-Complete problem, but this paper describes a practical method that performs provably close to the optimal. Interestingly they can optimize for performance or power efficiency, so you can use different algorithm
6 0.73147845 371 high scalability-2008-08-24-A Scalable, Commodity Data Center Network Architecture
7 0.72187412 374 high scalability-2008-08-30-Paper: GargantuanComputing—GRIDs and P2P
8 0.71589613 1316 high scalability-2012-09-04-Changing Architectures: New Datacenter Networks Will Set Your Code and Data Free
9 0.71456039 1039 high scalability-2011-05-12-Paper: Mind the Gap: Reconnecting Architecture and OS Research
10 0.68460089 1319 high scalability-2012-09-10-Russ’ 10 Ingredient Recipe for Making 1 Million TPS on $5K Hardware
11 0.68442988 1572 high scalability-2014-01-03-Stuff The Internet Says On Scalability For January 3rd, 2014
12 0.67220384 604 high scalability-2009-05-20-Paper: Flux: An Adaptive Partitioning Operator for Continuous Query Systems
13 0.67152882 1214 high scalability-2012-03-23-Stuff The Internet Says On Scalability For March 23, 2012
14 0.66513389 1116 high scalability-2011-09-15-Paper: It's Time for Low Latency - Inventing the 1 Microsecond Datacenter
15 0.66125441 1048 high scalability-2011-05-27-Stuff The Internet Says On Scalability For May 27, 2011
16 0.66124809 1339 high scalability-2012-10-12-Stuff The Internet Says On Scalability For October 12, 2012
17 0.65349859 1256 high scalability-2012-06-04-OpenFlow-SDN is Not a Silver Bullet for Network Scalability
18 0.64962435 1234 high scalability-2012-04-26-Akaros - an open source operating system for manycore architectures
19 0.64812911 146 high scalability-2007-11-08-scaling drupal - an open-source infrastructure for high-traffic drupal sites
20 0.64644724 1456 high scalability-2013-05-13-The Secret to 10 Million Concurrent Connections -The Kernel is the Problem, Not the Solution
topicId topicWeight
[(1, 0.123), (2, 0.169), (10, 0.034), (17, 0.027), (27, 0.026), (30, 0.032), (34, 0.152), (47, 0.012), (61, 0.055), (73, 0.014), (77, 0.035), (79, 0.143), (85, 0.055), (94, 0.045)]
simIndex simValue blogId blogTitle
1 0.93513203 777 high scalability-2010-02-15-Scaling Ambition at StackOverflow
Introduction: Joel Spolsky and Jeff Atwood are raising VC money for StackOverflow. This is interesting for three reasons: 1) Joel has always seemed like a keep it small and grow organically type of guy, so this is a big step in a different direction. 2) It means they think there's a very big market in the Q&A; space and they mean to capture as much as the market as possible. 3) Most importantly for this blog, Joel gives some good advice on when to stay fresh and local and when it's time to jump for the brass ring, scale up your ambition, and go for VC money. Please see Joel's blog post for the details, but here's when to go VC: There’s a land grab going on. There is a provable concept that’s repeatable. The business itself could benefit from the publicity. The investor will add substantial value to the business. The business can potentially have a big exit or become a large, publically traded company. The founders are not in it for their own personal aggrandizement. Joel t
same-blog 2 0.92427754 1594 high scalability-2014-02-12-Paper: Network Stack Specialization for Performance
Introduction: In the scalability is specialization department here is an interesting paper presented at HotNets '13 on high performance networking: Network Stack Specialization for Performance . The idea is generalizing a service so it fits in the kernel comes at a high performance cost. So move TCP into user space. The result is a web server with ~3.5x the throughput of Nginx "while experiencing low CPU utilization, linear scaling on multicore systems, and saturating current NIC hardware." Here's a good description of the paper published on Layer 9 : Traditionally, servers and OSes have been built to be general purpose. However now we have a high degree of specialization. In fact, in a big web service, you might have thousands of machines dedicated to one function. Therefore, there's scope for specialization. This paper looks at a specific opportunity in that space. Network stacks today are good for high throughput with large transfers, but not small files (which are common in web browsi
3 0.92096633 830 high scalability-2010-05-25-Strategy: Rule of 3 Admins to Save Your Sanity
Introduction: The idea came up in this Hacker News thread , commenting on a 37signals interview, that having three system administrators is the minimum optimal number of admins. Everyone wants to lower their costs by having each admin administer a lot of machines. The problem is when you have fewer than three admins you can never get a break from the constant corrosive pressure of always being on call. When every moment of your life you are dreading the next emergency, it eats at you. Having three admins solves that problem. With three admins you can: Go on a real vacation. The two remaining admins can switch off being on call. Not be on call all the time. A larger shop will naturally have more admins so it's not as big an issue, but at smaller shops trying to minimize head count, carrying three admins (or people in those roles) might be something to consider.
4 0.90308785 1092 high scalability-2011-08-04-Jim Starkey is Creating a Brave New World by Rethinking Databases for the Cloud
Introduction: Jim Starkey , founder of NuoDB , in this thread on the Cloud Computing group, delivers a masterful post on why he thinks the relational model is the best overall compromise amongst the different options, why NewSQL can free itself from the limitations of legacy SQL architectures, and how this creates a brave new lock free world.... I'll [Jim Starkey] go into more detail later in the post for those who care, but the executive summary goes like this: Network latency is relatively high and human attention span is relatively low. So human facing computer systems have to perform their work in a small number of trips between the client and the database server. But the human condition leads inexorably to data complexity. There are really only two strategies to manage this problem. One is to use coarse granularity storage, glombing together related data into a single blob and letting intelligence on the client make sense of it. The other is storing fine granularity data on the s
5 0.88595057 387 high scalability-2008-09-22-Paper: On Delivering Embarrassingly Distributed Cloud Services
Introduction: How do we scale datacenters? Should we build a few mammoth million machine datacenters or many smaller micro datacenters? Intuitively we usually go with a bigger is better economies of scale type argument, but it may not be so. What works for Walmart may not work for White Box World. Mega datacenters may actually exhibit diseconomies of scale. It may be better to run applications over many distributed micro datacenters instead of one large one. This paper by Ken Church, Albert Greenberg, and James Hamilton, all from Microsoft, takes a look at the different issues and concludes: Putting it all together, the micro model offers a design point with attractive performance, reliability, scale and cost. Given how much the industry is currently investing in the mega model, the industry would do well to consider the micro alternative. Related Articles Embarrasingly Distributed Cloud Services by James Hamilton Diseconomies of Scale by James Hamilton. Architecture
6 0.86997247 114 high scalability-2007-10-07-Product: Wackamole
7 0.85361576 765 high scalability-2010-01-25-Let's Welcome our Neo-Feudal Overlords
10 0.85010582 1389 high scalability-2013-01-18-Stuff The Internet Says On Scalability For January 18, 2013
11 0.8489787 1557 high scalability-2013-12-02-Evolution of Bazaarvoice’s Architecture to 500M Unique Users Per Month
12 0.84848148 1015 high scalability-2011-04-01-Stuff The Internet Says On Scalability For April 1, 2011
13 0.84660858 1275 high scalability-2012-07-02-C is for Compute - Google Compute Engine (GCE)
14 0.84620768 1630 high scalability-2014-04-11-Stuff The Internet Says On Scalability For April 11th, 2014
15 0.84605706 761 high scalability-2010-01-17-Applications Become Black Boxes Using Markets to Scale and Control Costs
16 0.84584624 666 high scalability-2009-07-30-Learn How to Think at Scale
17 0.84568214 1378 high scalability-2012-12-28-Stuff The Internet Says On Scalability For December 28, 2012
18 0.84558004 1612 high scalability-2014-03-14-Stuff The Internet Says On Scalability For March 14th, 2014
19 0.84522599 716 high scalability-2009-10-06-Building a Unique Data Warehouse
20 0.84518516 1170 high scalability-2012-01-06-Stuff The Internet Says On Scalability For January 6, 2012