Example: confidence

Cloud Computing with Microsoft Azure - Reliable …

Copyright 2009 Reliable Software, Computing with Microsoft AzureMichael Three FlavorsAzure Operating System (Platform as a Service)Worker/Web Role, Blobs, Queues, TablesAzure .NET Services (Software as a Service)Access Control ServiceSQL Azure (SQL Server in the sky)Workflow ServicesAzure Hosted Services (Application as a Service)Hosted ExchangeHost SharePoint Copyright 2009 Reliable Software, Operating SystemAbstracts the underlying infrastructureManages resources Copyright 2009 Reliable Software, PlatformService managementComputeStorageDeveloper experienceYou define rules and provide codePlatform deploys, monitors, and manages your service according to your rules Copyright 2009 Reliable Software, Services Platform Copyright 2009 Reliable Software, ServiceWeb/ Worker RoleBlobs, Tables, QueuesSQL AzureLive ServicesDynamic CRM ServicesAccess Control Demo.

Latency Exists Speed of light in fiber optic cable: 124,000 miles per second A ping Japan from Boston takes 100 ms.

Tags:

  Computing, With, Cloud, Microsoft, Azure, Cloud computing with microsoft azure

Information

Domain:

Source:

Link to this page:

Please notify us if you found a problem with this document:

Other abuse

Transcription of Cloud Computing with Microsoft Azure - Reliable …

1 Copyright 2009 Reliable Software, Computing with Microsoft AzureMichael Three FlavorsAzure Operating System (Platform as a Service)Worker/Web Role, Blobs, Queues, TablesAzure .NET Services (Software as a Service)Access Control ServiceSQL Azure (SQL Server in the sky)Workflow ServicesAzure Hosted Services (Application as a Service)Hosted ExchangeHost SharePoint Copyright 2009 Reliable Software, Operating SystemAbstracts the underlying infrastructureManages resources Copyright 2009 Reliable Software, PlatformService managementComputeStorageDeveloper experienceYou define rules and provide codePlatform deploys, monitors, and manages your service according to your rules Copyright 2009 Reliable Software, Services Platform Copyright 2009 Reliable Software, ServiceWeb/ Worker RoleBlobs, Tables, QueuesSQL AzureLive ServicesDynamic CRM ServicesAccess Control Demo.

2 Scalable ArchitectureCloud Storage (blob, table, queue)Web RoleLBnWorker RolemCloud Computing is Utility Computing Copyright 2009 Reliable Software, of Infinite Computing Resources on DemandNoup front commitmentPay for resources as neededThreeClasses of VendorsScalability, Failover, RecoveryAmazonGoogle / Copyright 2009 Reliable Software, Computing ScenariosOutsource Your InfrastructureOccasional Need for Massive ComputationNo Need to Build to Peak CapacityCloud-BurstingSoftware as a ServiceData Close To Your Customer Internet Scale Copyright 2009 Reliable Software, ConditionsPricingService Level Agreement (SLA) Copyright 2009 Reliable Software, Platform PricingCompute $ per hour Storage $ per GB month Storage Transactions $ per 10K Bandwidth $ in per GB $ out per GB Within the datacenter is free Copyright 2009 Reliable Software, AzureUp to 1 GB database $ /month Up to 10 GB database $ / month Bandwidth in per GB out per GB Copyright 2009 Reliable Software, Data Costs10 GB SQL Database2 GB a month data in, 4 GB a month data out$ a monthA SAN can cost from $30-40,00025 year equivalentDoes not consider the cost of infrastructure employees or the software licenses.

3 Copyright 2009 Reliable Software, Compute Costs$1051 per year for one compute process with no idle time$ if you did a storage save every second$3600 per year 2 TB of disk storageAbout $5000 / yearEmployee and licensing costs not considered Copyright 2009 Reliable Software, Azure SLAC omputation: up timeSQL Azure : up time Copyright 2009 Reliable Software, SLA20072008 GoalActualGoalActualCalls Answered Within30 Seconds80% # ServiceInterruptions Per # Min WithoutPower Per Appointments Meters Read "on cycle"vs Cases Per 1000 Copyright 2009 Reliable Software, Availability: CaseSMB ApplicationsMassive Computation NeedsNo Need to Build to Peak CapacityCloud BurstingSoftware as a Service Copyright 2009 Reliable Software, Computing ScenariosOutsource Your InfrastructureOccasional Need for Massive ComputationNo Need to Build to Peak CapacityCloud-BurstingSoftware as a ServiceData Close To Your Customer Internet Scale Copyright 2009 Reliable Software, ExistsSpeed of light in fiber optic cable: 124,000 miles per secondA ping Japan from Boston takes 100 number is about 250 10 images for a web site: 1 secondIgnores Latency of the operation Copyright 2009 Reliable Software, is LimitedShannon's Law.

4 C = B log2(1 + S / N)Capacity = bit / secondBandwidth (hertz)S/N * 5 to double capacity given bandwidth Copyright 2009 Reliable Software, is Not BandwidthSize of the shovel vs. how fast you can shovelInfinite shovel capacity(bandwidth) is limited by how fast one can shovel (latency). Copyright 2009 Reliable Software, Bandwidth, Poor LatencyBuy a two terabyte disk drivePut it in a car and drive to New York Copyright 2009 Reliable Software, to Move DataComputational Power Gets Cheaper Faster than Network BandwidthCheaper to compute where data is instead of moving itDistributed Computing Economics Jim Gray Want data to be close to where your customer is Copyright 2009 Reliable Software, is Not Always AvailableCell phoneData Center OutagesEquipment UpgradesData redundancy to improve reliability Copyright 2009 Reliable Software.

5 For Data Slows ComputationPartition Your Data to Improve PerformancePartition Your Data to Achieve Internet ScaleData Naturally Lives In Multiple PlacesDistributed Transactions Impede ThroughputHuman Interaction Copyright 2009 Reliable Software, Databases Scale Up Not OutRelational Databases scale well on a single node or clusterComplexity of relationsQuery plans with hundreds of options the query analyzer evaluates at runtimeNormalizationACID TransactionsTwo Phase Commit guarantees consistency if you have infinite timeQuick scale up difficult with hardware upgrade Copyright 2009 Reliable Software, Dictate Scale Out Not UpCheap, commodity hardware argues for spreading load across multiple serversHow do you distribute data among several databases?How do you achieve consistency?How do you achieve throughput with distributed transactions?

6 Copyright 2009 Reliable Software, TheoremConsistencyAvailabilityTolerance to network PartitionsCan Have Any TwoEric Brewer, UC Berkeley, Founder ~brewer/cs262b-2004 site DatabaseDatabase ClusterLDAPTwo phase commitValidate CacheConsistency and AvailabilityConsistencyAvailabilityParti tioningDistributed DatabaseDistributed LockingPessimistic LockingMinority Partitions InvalidConsistency and PartitioningConsistencyAvailabilityParti tioningForfeit ConsistencyGoogle BigTableAmazon Simple DBOptimisticCan DenormalizeNo ACID transactionsCompensationAvailability and PartitioningConsistencyAvailabilityParti tioningStorage in AzureWorld of ConsistencySQL AzureWorld of Internet Scale (Numbers or Geography)Blobs, Tables, Queues Copyright 2009 Reliable Software, to be SQL Server in the skyTables, Stored Procedures, Triggers, Constraints Views, IndicesUses TDS (Tabular Data Stream) ProtocolChange connection string to get to another SQL ServerNo Current Availability Get Started with SQL ExpressSQL Data ServicesTables of key/value pairs for highly scalable structured storageCRUD operationsNo FK relations, Joins, Constraints, SchemasPartition / Tables / Entities / PropertiesEntity has Unique Row KeyWindows Azure Storage ServicesFit well with tens or hundreds of commodity serversBetter mapping with objects than ORMNo integrity constraintsNo joined queriesNo standards among vendors (lock in)Will Microsoft have query limits?

7 Amazon no query longer than 5 secondsGoogle no more than 1000 items returnedAzure Storage Services Car TableKeyAttribute 1 Attribute 2 Attribute3 Attribute 41 Make: BMWC olor:GreyYear 20032 Make: NissanColor :RedYellowYear: 2005 Transmission: Easytronic3 Plane: BoeingColor: BlueEngine: Rolls RoyceNo PartitioningNatural PartitioningPartitioning for AvailabilityDo You Need To Partition Your Data to Scale?If you have to partition to scale, how do you decide between availability and consistency ?AmazonAirline reservationsStock TradesDeposit of a Bank CheckDeleting a photo from Flickr or FacebookWhat is the Cost of an Apology?AuthenticationSAML tokens expireLaunching a nuclear weaponSometimes the cost is too highVendor drops the last crystal vaseCheck bouncesDouble-entry bookkeeping requires compensation at least 13thcenturyEventually make consistentBusinesses Apologize AnywaySoftware approximates the state of the worldIt makes the best guess possibleSometimes that is wrongOther computers might have other opinionsOverturn software myths of the past 25 of the Software !

8 = State of the WorldBusiness DecisionHow much does it cost to get it absolutely right?Computers can remember their guessesCan replicate to share guessesIt may be cheaper to forget, and reconcile laterHow consistent?Identify objects by unique key (partition key / row key)Objects can move when repartitioningCannot assume two objects remain on the same machineData might go offlineTransactions can only apply on per object basisDifferent computations might come to different conclusionsDefine message based workflows for ultimate reconciliation and replication of resultsDesign For Eventual ConsistencyDemos ? Copyright 2009 Reliable Software, Azure is about understanding Economics of cost and availabilityNeed for ScalabilityArchitectural ImplicationsDesign for Eventual ConsistencyRemember the 2 / 10 rule Copyright 2009 Reliable Software, Inc.