Introduction I often advocate that distributed systems are the most difficult to change the way you think. For the excitation of this change, I Michael Kors Berkley Logo Large Beige Clutches found some very useful reading material. as follows. Michael Kors Berkley Logo Large Black Clutches Thought Provokers something that makes you think about the way you design the essay. Not all things can rely on large servers, databases, and things to be resolved. ! Brewer et al Harvest, Yield and Scalable Tolerant Systems CAP principles applied in the real world, from On Designing and Deploying Internet Scale Services James HamiltonLatency Exists, Cope processing delays and architecture on the impact of Latency - the new web performance bottleneck content is not too new, but it is worth next concern The Perils of Good Abstractions build the perfect API / interface is very difficult Chaotic Perspectives large-scale system developers do not like everything - unpredictable, chaotic, some of concurrent Website Architecture from various large-scale Website scalable architecture article Data on the Outside versus Data on the Inside Pat hellandMemories, Guesses and Apologies Pat HellandSOA and Newton's Universe - Pat HellandBuilding on Quicksand - Pat HellandWhy Distributed Computing - Jim WaldoA Note on Distributed Computing - Waldo, Wollrath et al Stevey's Google Platforms Rant - Yegge Amazon SOA platform experience some related technology, but more interesting is that they create a Michael Kors Bedford Logo Monogram Large Black Totes culture and structure of a mating. A Conversation with Werner Vogels on Amazon transition to a service-based architecture coverage of Discipline and Focus on Amazon transition to a service-based architecture of an interview with another Vogels on ScalabilitySOA creates order out of chaos @ AmazonGoogle current field of distributed systems 'Rocket Science' (described abstruse learning) Michael Kors Bedford Gusset Medium Tan Crossbody Bags MapReduceChubby Lock ManagerGoogle File SystemBigTableData Management for Internet-Scale Single-Sign-OnDremel: Interactive Analysis of Web-Scale DatasetsLarge-scale Incremental Processing Using Distributed Transactions and NotificationsMegastore: Providing Scalable, Highly Available Storage for Interactive Services - implement clever design across data centers, low-latency paxos algorithm. Spanner - Google's scalable, multi-version, globally distributed and replicated databases. Photon - continuous data stream, fault-tolerant and expansion. Michael Kors Clutches Expansion is very difficult, .Mesa especially in the clock skew, high availability and distributed situation: Geo-Replicated, Near Real-Time, Scalable Data Warehousing Michael Kors Bedford Signature Large White Totes for storing Google Internet advertising business-related data critical measurement data warehouse systems. eBay is interesting that they abandoned most of J2EE, and uses a lot of database partitions. Meanwhile, take a look at their website upgrade tool. SD Forum 2006 Michael Kors Bedford Logo Monogram Large Coffee Totes Michael Kors Berkley Logo Large Beige Clutches Model Consistency Key to adapt to the environment of the system is to find the right trade-off consistency and availability. CAP Conjecture - consistency, availability, partition tolerance can not simultaneously satisfy Consistency, Availability, and Convergence - proved that in a typical system consistency possible upper bound. CAP Twelve Years Later: How the 'Rules' Have Changed - Eric Brewer in the original job description weigh extended Consistency and Availability - VogelsEventual Consistency - VogelsAvoiding Two-Phase Michael Kors Bedford Signature Large White Totes Commit - way Michael Kors Berkley Logo Large Beige Clutches to avoid two-phase commit 2PC or not 2PC, Wherefore Art Thou XA - Two-phase commit is not a silver bullet Life Michael Kors Bedford Signature Large White Totes Beyond Distributed Transactions - HellandIf you have too much data, then 'good enough' is good enough - NoSQL, future data Theory - Pat HellandStarbucks does not do two phase commit - in the act asynchronous mechanism You Can not Sacrifice Partition Tolerance - another CAP Description Optimistic Replication - Data from weak master copy some consistency of approach theory describes a wide variety of distributed system design important factor in papers. Distributed Michael Kors Bedford Gusset Medium Orange Crossbody Bags Computing Economics - Jim GrayRules of Thumb in Data Engineering - Jim Gray and Prashant ShenoyFallacies of Distributed Computing - Peter DeutschImpossibility of distributed consensus with one faulty process is also known as FLP [access requires an account or paid, free version here: here] Unreliable Failure . Detectors for Reliable Distributed Systems FLP method of processing problems Lamport Clocks - when each computer's clock is independent of time, how do you build a global view of time. The Byzantine Generals ProblemLazy Replication: Exploiting the Semantics of Distributed ServicesScalable Agreement - Towards Ordering as a ServiceScalable Eventually Consistent Counters over Unreliable Networks in an unreliable world, scalable count difficult. Languages and tools using specific technology to Michael Kors Bedford Large Orange Shoulder Bags build distributed systems. Programming Distributed Erlang Applications: Pitfalls and Recipes building reliable distributed applications and select the Erlang OTP or the problem is not just that simple. Infrastructure Principles of Robust Timing over the Internet, even such basic things commissioning, management clock is also very important. Storage Consistent Michael Kors Clutches Hashing and Random TreesAmazon's Dynamo Storage ServicePaxos consistency algorithm to understand this algorithm is a challenge. I suggest that before reading other papers to read 'Paxos Michael Kors Outlet Made Simple', and then after reading other papers, read it again. The Part-Time Parliament - Leslie LamportPaxos Made Simple - Leslie LamportPaxos Made Live - An Engineering Perspective - Chandra et al Revisiting the Paxos Algorithm - Lynch et al How to build a highly available system with consensus - Butler LampsonReconfiguring a State Machine - Lamport et al. - Members of change cluster Implementing Fault-Tolerant Services Using the State Machine Approach: a Tutorial - Fred Schneider consistency other articles Mencius: Building Efficient Replicated State Machines for WANs - consistency algorithm for WAN Gossip Protocol (contagious behavior) Epidemic Routing BibliographyHow robust are gossip-based communication protocols Astrolabe: A Robust and Scalable Technology For Distributed Systems Monitoring, Management, and Data MiningEpidemic Computing at CornellFighting Fire With Fire: Using Randomized Gossip To Combat Stochastic Scalability LimitsBi-Modal MulticastACM SIGOPS Operating Systems Review - Gossip-based computer networkingSWIM: Scalable Weakly-consistent Infection-style Process Group Membership ProtocolP2PChord: Find a protocol for the Internet application point scalable. Kademlia: Scalable, decentralized object location and routing for large-scale point system: a point-based information system XOR Pastry. PAST: for large scale storage system SCRIBE persistent storage --Pastry on point: a large-scale and decentralized application layer multicast messaging system WAN infrastructure --Pastry on.distributed system Reading List