data liberty

44
Data Liberty Alternatives to the shackles of limited scale in data solutions Andy Cross Windows Azure MVP Elastacloud

Upload: charo

Post on 23-Feb-2016

43 views

Category:

Documents


0 download

DESCRIPTION

Data Liberty. Alternatives to the shackles of limited scale in data solutions Andy Cross Windows Azure MVP Elastacloud. Social Media. Tell everyone I’m awesome #cloudburst I’m @ andybareweb. Data value at scale requires technology choices;. - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Data Liberty

Data LibertyAlternatives to the shackles of limited scale in data

solutions

Andy CrossWindows Azure MVP

Elastacloud

Page 2: Data Liberty

Tell everyone I’m awesome #cloudburst

I’m @andybareweb

Social Media

Page 3: Data Liberty

Data value at scale requires technology choices;• often prioritising data read traversal over operational

characteristics of create/update/delete

• embracing hybrid data platforms with varied technology partners over homogenous estates

• establishing alternative skillsets, augmented with entrenched languages, trusting cloud over maintenance

• following robust engineering processes to provide rigour in a deterministic world

Page 4: Data Liberty

Bravery leads to rewards;• the winners will have data which shows them that they’ve

won

• the commoditised query turns energy sucking data silos into profit centres

• new data traversal mechanisms lead to new connotative data expression

• everything you already know is relevant and valid; the constraints on how it is applied are not

Page 5: Data Liberty

Most developers have heard of Big Data. I’m going to show how Microsoft are increasingly relevant in this space.

My talk is about architecture and approach. Note we’re talking Big Data and not strictly Data Science.

But it’s always worth context so lets start with the history.

WHAT’S A DATA SCIENTISTS FAVOURITE LANGUAGE?

Page 6: Data Liberty

IBM have been a leader in Big Data for years.

Wikimedia commons

Page 7: Data Liberty

We’re not as great as we’d hope; we’re often still bound by our ability to marshal our IO.

Just as the speed of loading punchcards was historically a limiting factor, we are now limited by our capacity to ingest data on individual machines.

This leads to ideas such as DFS and data locality.

Page 8: Data Liberty
Page 9: Data Liberty

During the evolution of data we eventually moved to client/server and this was a big step up from dBase et al of the time.

Fundamentally however, the tabular structured nature of data poses many changes; not least the long term effects of normalisation which trade off effective storage in the short term with long term offset compute which is required to reconstruct sets.

This eventually leads to such ideas as NoSQL document and entity stores.

Page 10: Data Liberty
Page 11: Data Liberty

Modelling of data provides a consistent challenge. Our world is highly connected and our brains are effective connectors of data. Real world data fits poorly into highly structured data sets.

This leads to semi-structured and unstructured data formats and data queryability through relationship traversal

Page 12: Data Liberty

The technologies shown today are primarily written in non-.net and non-Microsoft languages and frameworks. Every time we do this, I’ll show examples ONLY in the .net and Microsoft stacks.

There are obviously challenges beyond language to running the alternative stacks; but remember in the Cloud you aren’t responsible for tuning a Linux cluster which has been running for 5 years. You should provision for a duration that is bounded by the likelihood of the cluster requiring routine maintenance.

Page 13: Data Liberty

Open Source; Apache Foundation.

Java.

Map Reduce framework for job distribution; Distributed File System for file access.

In Windows Azure this is known as HDInsight.

Hadoop – KEY FACTS

Page 14: Data Liberty

Hadoop is O(n)It exhibits linear performance; when the dataset doubles, the time taken to execute the algorithm doubles.

Page 15: Data Liberty

Lets look at some scary JavaAny children should look away now.

Page 16: Data Liberty
Page 17: Data Liberty
Page 18: Data Liberty

Hadoop SDK

C# integrationRemote Data & JobsHive in C#Serialization

Page 19: Data Liberty

public class SwedishSessionsJob : HadoopJob<SwedishSessionsMapper, SessionsReducer> { public override HadoopJobConfiguration Configure(ExecutorContext context) { var config = new HadoopJobConfiguration() { InputPath = "\"/AllSessions/*.gz\"", OutputFolder = "/SwedishSessions/" }; return config; } }

Jobs

Page 20: Data Liberty

public class SwedishSessionsMapper : MapperBase { public override void Map(string inputLine, MapperContext context) { if (inputLine.Contains("Country=Sweden") { context.IncrementCounter("SwedishSession"); context.EmitKeyValue(“SE", "1"); } } }

Mapper

Page 21: Data Liberty

public class SessionsReducer : ReducerCombinerBase { public override void Reduce(string key, IEnumerable<string> values, ReducerContext context) { context.EmitKeyValue(key, values.Count()); } }

Reducer

Page 22: Data Liberty

Testing Hadoop Queries

var inputData = "Country=Sweden&Name=Magnus";var result = StreamingUnit.Execute<Jobs.SwedishJob>(new[]{inputData});Assert.AreEqual("SE\t1", result.ReducerResult.First());

Page 23: Data Liberty

Skill reuse

Express elegant solutions in C#Familiar Unit Testing patternsConcise programmatic terseness

Your existing development team can immediately

realise value

The frameworks

facilitate deterministic

testing for highly reliable

queries

Complex logic is best expressed in programmatic

form

Page 24: Data Liberty

Commoditised query

Provision

Execute

De-provision

Valu

e

Action Cost

Value of query

Time

Cost

Page 25: Data Liberty

* Tools are great but not friendly

HDInsight wins.Automated provisioning and job execution services.

Transient clusters limit exposure to poorly tooled* java estate.

Persistence with Windows Azure Blob Storage as HDFS proxy known as Azure Storage Vault (ASV).

Persistence in Windows Azure SQL Database for Hive Metastore.

Javascript console.

Page 26: Data Liberty

NoSQL Document and Entity StoresExamples in MongoDB

Entity stores are similar; you can find a great example in Windows Azure Table Storage

Page 27: Data Liberty

What is a document database?Relational Database Document Database

{ "_id" : ObjectId("51fccc57f82352d76653bdae"), "Name" : { "FirstName" : "Owen", "LastName" : "Grzegorek" }, "Company" : "Howard Miller Co", "Address" : { "Line1" : "15410 Minnetonka Industrial Rd", "Line2" : "Minnetonka", "Line3" : "Hennepin", "Line4" : "MN", "Line5" : "55345" }, "ContactDetails" : { "Phone" : "952-939-2973", "Fax" : "952-939-4663", "Email" : "[email protected]", "Web" : "http://www.owengrzegorek.com" }}

{ "_id" : ObjectId("51fccc57f82352d76653bdae"), "Name" : { "FirstName" : "Owen", "LastName" : "Grzegorek" }, "Company" : "Howard Miller Co", "Address" : { "Line1" : "15410 Minnetonka Industrial Rd", "Line2" : "Minnetonka", "Line3" : "Hennepin", "Line4" : "MN", "Line5" : "55345" }, "ContactDetails" : { "Phone" : "952-939-2973", "Fax" : "952-939-4663", "Email" : "[email protected]", "Web" : "http://www.owengrzegorek.com" }}

{ "_id" : ObjectId("51fccc57f82352d76653bdae"), "Name" : { "FirstName" : "Owen", "LastName" : "Grzegorek" }, "Company" : "Howard Miller Co", "Address" : { "Line1" : "15410 Minnetonka Industrial Rd", "Line2" : "Minnetonka", "Line3" : "Hennepin", "Line4" : "MN", "Line5" : "55345" }, "ContactDetails" : { "Phone" : "952-939-2973", "Fax" : "952-939-4663", "Email" : "[email protected]", "Web" : "http://www.owengrzegorek.com" }}

{ "_id" : ObjectId("51fccc57f82352d76653bdae"), "Name" : { "FirstName" : "Owen", "LastName" : "Grzegorek" }, "Company" : "Howard Miller Co", "Address" : { "Line1" : "15410 Minnetonka Industrial Rd", "Line2" : "Minnetonka", "Line3" : "Hennepin", "Line4" : "MN", "Line5" : "55345" }, "ContactDetails" : { "Phone" : "952-939-2973", "Fax" : "952-939-4663", "Email" : "[email protected]", "Web" : "http://www.owengrzegorek.com" }}

{ "Name" : { "FirstName" : "Owen", "LastName" : "Grzegorek" }, "Company" : "Howard Miller Co", "Address" : { "Line1" : "15410 Minnetonka Industrial Rd", "Line2" : "Minnetonka", "Line3" : "Hennepin", "Line4" : "MN", "Line5" : "55345" }, "ContactDetails" : { "Phone" : "952-939-2973", "Fax" : "952-939-4663", "Email" : "[email protected]", "Web" : "http://www.owengrzegorek.com" }}

{ "Name" : { "FirstName" : "Owen", "LastName" : "Grzegorek" }, "Company" : "Howard Miller Co", "Address" : { "Line1" : "15410 Minnetonka Industrial Rd", "Line2" : "Minnetonka", "Line3" : "Hennepin", "Line4" : "MN", "Line5" : "55345" }, "ContactDetails" : { "Phone" : "952-939-2973", "Fax" : "952-939-4663", "Email" : "[email protected]", "Web" : "http://www.owengrzegorek.com" }}

{ "Name" : “Richard Conway", “Books Published” : “12”, “Specialises in” : “Data Science”}

{ "Name" : “Andy Cross", “Hometown" : “Blackpool“}

{ "Name" : “Isaac Abraham", “Age" : “33“ “Football Team” : “Tottenham” “Icon” : }

Page 28: Data Liberty

MongoDB Key Facts• General Purpose Operational Database• Real-time updates, ad-hoc queries and batch processing• Maps nicely with popular programming models e.g. .NET• Schema-free documents – lightweight and quick to get up and running

• High Performance• Embedding documents – no expensive joins across tables• Indexes allow query optimization• High-speed saving of data (writes)

• High Availability• Built in replication• Built in failover

• Easy Scalability• “Sharding” allows easily spreading data across multiple databases• Replicated data can be spread throughout the cluster

Page 29: Data Liberty

MongoDB is O(log n)It exhibits logarithmic performance; when the dataset doubles, the time taken to execute the algorithm increases by a fixed amount

Page 30: Data Liberty

Strengths of MongoDBLow barrier to entryUses well-known .NET technologies e.g. LINQGood migration path from SQL-style development

Good fit for .NET developers

Works well as operational data storeBatch processing capability for map reduceFlexible

Massively scalable with well-defined replication modelSelf-managing – easily add new nodesHigh performance writes and eventually consistent reads

Designed for scalability

Database is free to use (tooling is not!)Popular, so a relatively large communityLow cost

Page 31: Data Liberty

Mongo SDK

There are many different way to connect with MongoDB from a .net project.

Official

Wrapper

Alternative

Tool

Page 32: Data Liberty

C# implementationsIf your data is regularly structured, you can use domain classes:

public class Book { public string Author { get; set; } public string Title { get; set; } }// "entities" is the name of the collection

var books = database.GetCollection<Entity>("books"); Book book = new Book

{ Author = "Ernest Hemingway", Title = "For Whom the Bell Tolls" }; books.Insert(book);

Page 33: Data Liberty

C# implementationsIf your data is irregularly structured or semi-structured, you can use a BSON object model:

BsonDocument person = new BsonDocument { { "name", "John Doe" }, { "address", new BsonDocument { { "street", "123 Main St." }, { "city", "Centerville" }, { "state", "PA" }, { "zip", 12345} }}};var people = database.GetCollection<BsonDocument>("people");people.Insert(person);

Page 34: Data Liberty

NoSQL Document WinsSemi-structured data first class citizen

Built in MapReduce

Operational and interactive

Massively scalable

Page 35: Data Liberty

Graph Databases, Neo4j KEY FACTSOpen Source; Neotechnologies

Java

Runs equally well on Windows or Linux. In Windows Azure there are VMDepot images able to be deployed in a few simple steps. Additionally the Azure Linux VMs are a good fit for this database engine.

There is an Open Source .net SDK available through Nuget and actively maintained primarily by an Australian company, Readify.

Page 36: Data Liberty

Neo4j is O(1)It exhibits constant-time performance; that is, the algorithm takes the same time to execute irrespective of the size of the dataset.

Page 37: Data Liberty

How O(1)?• Graphs don’t have tables. They don’t have collections.• They have nodes and relationships.

• Rather than having to select out a whole table, we can identify a point on the graph• A start point

• Follow the traversal of relationships from that point.

Page 38: Data Liberty

http://www.apcjones.com/arrows/#

Page 39: Data Liberty

Things we can do• Find all the things formed in Sweden START sweden = node:countryIdx(“country=Sweden”)MATCH Sweden<-[:FORMED_IN]-somethingRETURN something;• Find friends of friendsSTART magnus = node:peopleIdx(“name=magnus”)MATCH magnus-[:FRIENDS]->friend-[:FRIENDS]->friendoffriendRETURN friendoffriend;

Page 40: Data Liberty

NEO4J Client

Open source Neo4j Client

Page 41: Data Liberty

C# examples

var query = neo4Jclient.Cypher .Start(new { sweden = Node.ByIndexLookup("countryIdx", "country", "sweden") }) .Match("sweden-[:FRIENDS]->friend-[:FRIENDS]->friendoffriend") .Return<Node<Friend>>("friendoffriend");

Page 42: Data Liberty

Graph Database Wins• Modelled domains match cognitive processes

• Optimised for traversal of relationships allow complex and “social” queries to emerge• LIKES of FRIENDS of COLLEAGUES

• O(1) performance characteristics due to ability to START queries at arbitrary graph points.

Page 43: Data Liberty

Summary

•HDInsight brings Hadoop to Azure•Suited to Data Volume, Variety, Variability etc•MongoDB brings Document stores•Suited to Data Volume, Operational concerns•Neo4j brings Graph database•Suited to data relationship traversal