NoSQL Landscape: Speed, Scale, and JSON

@jchris Chris Anderson NoSQL Landscape & Grid Compu7ng 1 Saturday,
October 6, 12

• 2.2 Billion internet users • 50% Americans use
smartphones • Your app can grow overnight • Are you ready? 2 Growth is the New Reality Saturday, October 6, 12

Instagrowth: Android Launch • Instagram gained nearly 1 million users
overnight when they expanded to Android Example 3 Saturday, October 6, 12

1 Instagram = 7.5M MAU* 4 Instagrowth: Android Launch Example
Saturday, October 6, 12

Draw Something -‐ Social Game 5 35 million monthly active
users in 1 month about 5 Instagrams (Instagram today is waaaay more than 1 Instagram) Saturday, October 6, 12

Goes Viral 3 Weeks aOer Launch 6 19 17 15
13 11 9 7 5 3 3/1 28 26 24 22 20 18 16 14 12 10 8 2/6 Draw Something by OMGPOP Daily Ac)ve Users (millions) 21 2 4 6 8 10 12 14 16 35+M MAU at peak Saturday, October 6, 12

By Contrast, at 1/2 an Instagram 7 The Simpson’s: Tapped
Out Daily Ac)ve Users (millions) Saturday, October 6, 12

GETTING IT RIGHT 8 Saturday, October 6, 12

Scalable Data Layer 9 •On-‐demand cluster sizing •Grow or shrink
with workload •Easy node provisioning •All nodes are the same •MulA-‐master Cross-‐Datacenter ReplicaAon •For a fast and reliable user experience worldwide •EﬀecAve Auto-‐sharding •Should avoid cluster hot spots Saturday, October 6, 12

Old School Hits a Scale Wall 10 Application Scales Out
Just add more commodity web servers Database Scales Up Get a bigger, more complex server Expensive & disruptive sharding, doesn’t perform at web scale Saturday, October 6, 12

Tradi^onal MySQL + Memcached Architecture 11 • Run as many
MySQL machines as you need • Data sharded evenly across the machines using client code • Memcached used to provide faster response time for users and reduce load on the database Memcached Tier MySQL Tier App Servers www.example.com Saturday, October 6, 12

Limitaôns of MySQL + Memcached 12 • To scale you
need to start using MySQL more simply • Scale by hand • Replication / Sharding is a black art • Code overhead to manage keeping memcache and mysql in sync • Lots of components to deploy Learn From Others -‐ This Scenario Costs Time and Money. Scaling SQL is potenâlly disastrous when going Viral: very risky ^me for major code changes and migraôns... you have no Time when skyrocke^ng up. Saturday, October 6, 12

NoSQL Architectural Promise 13 Couchbase Database Servers App Servers www.example.com
• High Performance data access • Scale Up/Down Horizontally • 24x7x365 Always-On Availability • Flexible Schema Document Model Saturday, October 6, 12

14 NOSQL TAXONOMY Saturday, October 6, 12

15 The Key-Value Store – the foundation of NoSQL Key
101100101000100010011101 101100101000100010011101 101100101000100010011101 101100101000100010011101 101100101000100010011101 101100101000100010011101 101100101000100010011101 101100101000100010011101 101100101000100010011101 101100101000100010011101 101100101000100010011101 101100101000100010011101 101100101000100010011101 101100101000100010011101 101100101000100010011101 Opaque Binary Value Saturday, October 6, 12

16 Memcached – the NoSQL precursor Key 101100101000100010011101 101100101000100010011101 101100101000100010011101
101100101000100010011101 101100101000100010011101 101100101000100010011101 101100101000100010011101 101100101000100010011101 101100101000100010011101 101100101000100010011101 101100101000100010011101 101100101000100010011101 101100101000100010011101 101100101000100010011101 101100101000100010011101 Opaque Binary Value memcached In-‐memory only Limited set of opera^ons Blob Storage: Set, Add, Replace, CAS Retrieval: Get Structured Data: Append, Increment “Simple and fast.” Challenges: cold cache, disrup^ve elas^city Saturday, October 6, 12

17 Redis – More “Structured Data” commands Key 101100101000100010011101 101100101000100010011101
101100101000100010011101 101100101000100010011101 101100101000100010011101 101100101000100010011101 101100101000100010011101 101100101000100010011101 101100101000100010011101 101100101000100010011101 101100101000100010011101 101100101000100010011101 101100101000100010011101 101100101000100010011101 101100101000100010011101 “Data Structures” Blob List Set Hash … redis In-‐memory only Vast set of opera^ons Blob Storage: Set, Add, Replace, CAS Retrieval: Get, Pub-‐Sub Structured Data: Strings, Hashes, Lists, Sets, Sorted lists Example opera7ons for a Set Add, count, subtract sets, intersec^on, is member?, atomic move from one set to another Saturday, October 6, 12

18 NoSQL catalog Key-‐Value memcached redis Data Structure Document Column
Graph Cache (memory only) Saturday, October 6, 12

19 Membase – From key-‐value cache to database Disk-‐based with
built-‐in memcached cache Cache reﬁll on restart Memcached compa^ble (drop in replacement) Highly-‐available (data replica^on) Add or remove capacity to live cluster “Simple, fast, elas^c.” membase Key 101100101000100010011101 101100101000100010011101 101100101000100010011101 101100101000100010011101 101100101000100010011101 101100101000100010011101 101100101000100010011101 101100101000100010011101 101100101000100010011101 101100101000100010011101 101100101000100010011101 101100101000100010011101 101100101000100010011101 101100101000100010011101 101100101000100010011101 Opaque Binary Value Saturday, October 6, 12

20 NoSQL catalog Key-‐Value memcached membase redis Data Structure Document
Column Graph Cache (memory only) Database (memory/disk) Saturday, October 6, 12

21 Couchbase – document-‐oriented database Key {
“string” : “string”, “string” : value, “string” : { “string” : “string”, “string” : value }, “string” : [ array ] } Auto-‐sharding Disk-‐based with built-‐in memcached cache Cache reﬁll on restart Memcached compa^ble (drop in replace) Highly-‐available (data replica^on) Add or remove capacity to live cluster When values are JSON objects (“documents”): Create indices, views and query against the views JSON OBJECT (“DOCUMENT”) Couchbase Saturday, October 6, 12

Column Graph Cache (memory only) Database (memory/disk) membase couchbase Saturday, October 6, 12

23 MongoDB – Document-‐oriented database Key {
“string” : “string”, “string” : value, “string” : { “string” : “string”, “string” : value }, “string” : [ array ] } Disk-‐based with in-‐memory “caching” BSON (“binary JSON”) format and wire protocol Master-‐slave replica^on Auto-‐sharding Values are BSON objects Supports ad hoc queries – best when indexed BSON OBJECT (“DOCUMENT”) MongoDB Saturday, October 6, 12

Column Graph mongoDB couchbase Cache (memory only) Database (memory/disk) Saturday, October 6, 12

25 Cassandra – Column overlays Disk-‐based system Clustered External
caching required for low-‐latency reads “Columns” are overlaid on the data Not all rows must have all columns Supports eﬃcient queries on columns Restart required when adding columns Good cross-‐datacenter support Cassandra Column 1 Column 2 Column 3 (not present) Saturday, October 6, 12

Column Graph mongoDB couchbase cassandra Cache (memory only) Database (memory/disk) Saturday, October 6, 12

27 Neo4j – Graph database Disk-‐based system External caching required
for low-‐latency reads Nodes, rela^onships and paths Proper^es on nodes Delete, Insert, Traverse, etc. Neo4j Saturday, October 6, 12

Column Graph mongoDB couchbase cassandra Cache (memory only) Database (memory/disk) Neo4j Saturday, October 6, 12

The Landscape 29 Speed Scale Couchbase Redis S3 Cassandra MongoDB
Riak HBase CouchDB Neo4j SimpleDB memcached RDBMS Datomic Saturday, October 6, 12

Datomic -‐ immutable func^onal data 30 Saturday, October 6, 12

Hello Couchbase Server 2.0 31 Saturday, October 6, 12

Couchbase Server 2.0 beta 32 Saturday, October 6, 12

33 Couchbase handles real world scale Saturday, October 6, 12

(Really) High Performance 34 Latency less than 1/2 ms Throughput
grows linearly with cluster size 5 Nodes -- 1.75M operations per second Cisco and Solarflare benchmark of Couchbase Server Saturday, October 6, 12

How fast? 35 hrp://www.slideshare.net/renatko/couchbase-‐performance-‐benchmarking Saturday, October 6, 12

Latency Saturday, October 6, 12

COMPLEXITY IS THE ENEMY 37 Saturday, October 6, 12

38 Couchbase Server Basic Opera^on COUCHBASE CLIENT LIBRARY §Docs distributed
evenly across servers in the cluster §Each server stores both ac)ve & replica docs § Only one server ac^ve at a ^me §Client library provides app with simple interface to database §Cluster map provides map to which server doc is on § App never needs to know § App reads, writes, updates docs § Mul^ple App Servers can access same document at same ^me Doc 2 Doc 5 SERVER 1 Doc 4 SERVER 2 Doc 1 SERVER 3 COUCHBASE CLIENT LIBRARY Doc 9 Doc 7 Doc 8 Doc 6 Doc 3 DOC DOC DOC DOC DOC DOC DOC DOC DOC Ac^ve Docs Ac^ve Docs Ac^ve Docs CLUSTER MAP CLUSTER MAP APP SERVER 1 APP SERVER 2 COUCHBASE SERVER CLUSTER Saturday, October 6, 12

evenly across servers in the cluster §Each server stores both ac)ve & replica docs § Only one server ac^ve at a ^me §Client library provides app with simple interface to database §Cluster map provides map to which server doc is on § App never needs to know § App reads, writes, updates docs § Mul^ple App Servers can access same document at same ^me Doc 4 Doc 2 Doc 5 SERVER 1 Doc 6 Doc 4 SERVER 2 Doc 7 Doc 1 SERVER 3 Doc 3 COUCHBASE CLIENT LIBRARY Doc 9 Doc 7 Doc 8 Doc 6 Doc 3 DOC DOC DOC DOC DOC DOC DOC DOC DOC DOC DOC DOC DOC DOC DOC Doc 9 Doc 5 DOC DOC DOC Doc 1 Doc 8 Doc 2 Replica Docs Replica Docs Replica Docs Ac^ve Docs Ac^ve Docs Ac^ve Docs CLUSTER MAP CLUSTER MAP APP SERVER 1 APP SERVER 2 COUCHBASE SERVER CLUSTER Saturday, October 6, 12

evenly across servers in the cluster §Each server stores both ac)ve & replica docs § Only one server ac^ve at a ^me §Client library provides app with simple interface to database §Cluster map provides map to which server doc is on § App never needs to know § App reads, writes, updates docs § Mul^ple App Servers can access same document at same ^me Doc 4 Doc 2 Doc 5 SERVER 1 Doc 6 Doc 4 SERVER 2 Doc 7 Doc 1 SERVER 3 Doc 3 Read/Write/Update COUCHBASE CLIENT LIBRARY Read/Write/Update Doc 9 Doc 7 Doc 8 Doc 6 Doc 3 DOC DOC DOC DOC DOC DOC DOC DOC DOC DOC DOC DOC DOC DOC DOC Doc 9 Doc 5 DOC DOC DOC Doc 1 Doc 8 Doc 2 Replica Docs Replica Docs Replica Docs Ac^ve Docs Ac^ve Docs Ac^ve Docs CLUSTER MAP CLUSTER MAP APP SERVER 1 APP SERVER 2 COUCHBASE SERVER CLUSTER Saturday, October 6, 12

39 Add Nodes to the Cluster § Two servers added
to cluster § One-‐click opera^on § Docs automa^cally rebalanced across cluster § Even distribu^on of docs § Minimum doc movement § Cluster map updated § App database calls now distributed over larger # of servers Doc 7 Doc 9 Doc 3 Ac^ve Docs Replica Docs Doc 6 COUCHBASE CLIENT LIBRARY CLUSTER MAP APP SERVER 1 COUCHBASE CLIENT LIBRARY CLUSTER MAP APP SERVER 2 Doc 4 Doc 2 Doc 5 SERVER 1 Doc 6 Doc 4 SERVER 2 Doc 7 Doc 1 SERVER 3 Doc 3 Doc 9 Doc 7 Doc 8 Doc 6 Doc 3 DOC DOC DOC DOC DOC DOC DOC DOC DOC DOC DOC DOC DOC DOC DOC Doc 9 Doc 5 DOC DOC DOC Doc 1 Doc 8 Doc 2 Replica Docs Replica Docs Replica Docs Ac^ve Docs Ac^ve Docs Ac^ve Docs COUCHBASE SERVER CLUSTER Saturday, October 6, 12

to cluster § One-‐click opera^on § Docs automa^cally rebalanced across cluster § Even distribu^on of docs § Minimum doc movement § Cluster map updated § App database calls now distributed over larger # of servers Doc 7 Doc 9 Doc 3 Ac^ve Docs Replica Docs Doc 6 COUCHBASE CLIENT LIBRARY CLUSTER MAP APP SERVER 1 COUCHBASE CLIENT LIBRARY CLUSTER MAP APP SERVER 2 Doc 4 Doc 2 Doc 5 SERVER 1 Doc 6 Doc 4 SERVER 2 Doc 7 Doc 1 SERVER 3 Doc 3 Doc 9 Doc 7 Doc 8 Doc 6 Doc 3 DOC DOC DOC DOC DOC DOC DOC DOC DOC DOC DOC DOC DOC DOC DOC Doc 9 Doc 5 DOC DOC DOC Doc 1 Doc 8 Doc 2 Replica Docs Replica Docs Replica Docs Ac^ve Docs Ac^ve Docs Ac^ve Docs SERVER 4 SERVER 5 Ac^ve Docs Ac^ve Docs Replica Docs Replica Docs COUCHBASE SERVER CLUSTER Saturday, October 6, 12

to cluster § One-‐click opera^on § Docs automa^cally rebalanced across cluster § Even distribu^on of docs § Minimum doc movement § Cluster map updated § App database calls now distributed over larger # of servers Read/Write/Update Read/Write/Update Doc 7 Doc 9 Doc 3 Ac^ve Docs Replica Docs Doc 6 COUCHBASE CLIENT LIBRARY CLUSTER MAP APP SERVER 1 COUCHBASE CLIENT LIBRARY CLUSTER MAP APP SERVER 2 Doc 4 Doc 2 Doc 5 SERVER 1 Doc 6 Doc 4 SERVER 2 Doc 7 Doc 1 SERVER 3 Doc 3 Doc 9 Doc 7 Doc 8 Doc 6 Doc 3 DOC DOC DOC DOC DOC DOC DOC DOC DOC DOC DOC DOC DOC DOC DOC Doc 9 Doc 5 DOC DOC DOC Doc 1 Doc 8 Doc 2 Replica Docs Replica Docs Replica Docs Ac^ve Docs Ac^ve Docs Ac^ve Docs SERVER 4 SERVER 5 Ac^ve Docs Ac^ve Docs Replica Docs Replica Docs COUCHBASE SERVER CLUSTER Saturday, October 6, 12

40 Fail Over Node COUCHBASE CLIENT LIBRARY CLUSTER MAP COUCHBASE
CLIENT LIBRARY CLUSTER MAP Doc 4 Doc 2 Doc 5 SERVER 1 Doc 6 Doc 4 SERVER 2 SERVER 3 Doc 3 Doc 9 Doc 7 Doc 8 Doc 6 DOC DOC DOC DOC DOC DOC DOC DOC DOC DOC DOC Doc 5 DOC DOC DOC Doc 1 Doc 8 Doc 2 Replica Docs Replica Docs Replica Docs Ac^ve Docs Ac^ve Docs Ac^ve Docs SERVER 4 SERVER 5 Ac^ve Docs Ac^ve Docs Replica Docs Replica Docs COUCHBASE SERVER CLUSTER § App servers happily accessing docs on Server 3 § Server fails § App server requests to server 3 fail § Cluster detects server has failed § Promotes replicas of docs to ac)ve § Updates cluster map § App server requests for docs now go to appropriate server § Typically rebalance would follow Doc 7 Doc 9 Doc 3 Ac^ve Docs Replica Docs Doc 6 Doc 7 Doc 1 Doc 3 DOC DOC Doc 9 DOC DOC Saturday, October 6, 12

40 Fail Over Node COUCHBASE CLIENT LIBRARY CLUSTER MAP COUCHBASE
CLIENT LIBRARY CLUSTER MAP Doc 4 Doc 2 Doc 5 SERVER 1 Doc 6 Doc 4 SERVER 2 SERVER 3 Doc 3 Doc 9 Doc 7 Doc 8 Doc 6 DOC DOC DOC DOC DOC DOC DOC DOC DOC DOC DOC Doc 5 DOC DOC DOC Doc 1 Doc 8 Doc 2 Replica Docs Replica Docs Replica Docs Ac^ve Docs Ac^ve Docs Ac^ve Docs SERVER 4 SERVER 5 Ac^ve Docs Ac^ve Docs Replica Docs Replica Docs COUCHBASE SERVER CLUSTER § App servers happily accessing docs on Server 3 § Server fails § App server requests to server 3 fail § Cluster detects server has failed § Promotes replicas of docs to ac)ve § Updates cluster map § App server requests for docs now go to appropriate server § Typically rebalance would follow Saturday, October 6, 12

41 •Suddenly, disk writes all began to time out •Many
services experienced outages: • FourSquare, Reddit, Quora, among others •With memory buffered writes, a scalable data layer keeps working • When EBS came back online, Couchbase wrote all the updated data to disk without missing a beat. War Story: EBS Outage Saturday, October 6, 12

42 Cross Data Center Replica^on §Data close to users §Mul^ple
loca^ons for disaster recovery §Independently managed clusters serving local data US DATA CENTER EUROPE DATA CENTER ASIA DATA CENTER Replica7on Replica7on Replica7on Saturday, October 6, 12

43 Built for Produc^on Saturday, October 6, 12

JSON DOCUMENT DATABASE 44 Saturday, October 6, 12

45 Document Database as Aggregate Database hrp://mar^nfowler.com/bliki/AggregateOrientedDatabase.html Saturday, October 6,
12

46 Document Database This synergy between the programming model and
the distribution model is very valuable. It allows the database to use its knowledge of how the application programmer clusters the data to help performance across the cluster. hrp://mar^nfowler.com/bliki/AggregateOrientedDatabase.html o::1001 { uid: ji22jd, customer: Ann, line_items: [ { sku: 0321293533, quan: 3, unit_price: 48.0 }, { sku: 0321601912, quan: 1, unit_price: 39.0 }, { sku: 0131495054, quan: 1, unit_price: 51.0 } ], payment: { type: Amex, expiry: 04/2001, last5: 12345 } } Saturday, October 6, 12

Developers <3 JSON 47 Saturday, October 6, 12

4 LET’S GET POST-‐RELATIONAL! Saturday, October 6, 12

49 JSON Documents • Maps more closely to external API
• CRUD Opera^ons, lightweight schema • Stored under an iden^ﬁer key { “fields” : [“with basic types”, 3.14159, true], “like” : “your favorite language” } client.set(“mydocumentid”, myDocument); mySavedDocument = client.get(“mydocumentid”); Saturday, October 6, 12

Meta + Document Body 50 { "brewery": "New
Belgium Brewing", "name": "1554 Enlightened Black Ale", "abv": 5.5, "descrip7on": "Born of a ﬂood...", "category": "Belgian and French Ale", "style": "Other Belgian-‐Style Ales", "updated": "2010-‐07-‐22 20:00:20" } { "id" : "beer_Enlightened_Black_Ale”, ... { Document user data, can be anything unique ID Metadata identiﬁer, expiration, etc “vintage” date format from an SQL dump >_< Saturday, October 6, 12

{ "brewery": "New Belgium "name":
"1554 Enlightene "abv": 5.5, "descrip7on": "Born of a ﬂ "category": "Belgian and F "style": "Other Belgian-‐Sty "updated": "2010-‐07-‐22 2 “ra7ngs” : { “525” : 5, “30” : 4, “1044” : 2 }, “comment_ids” : [ “f1e62”, “6ad8c” ] } Add comments to the beer { "type": "comment", "about_id": "beer_Enlightened_Black_Ale", "user_id": 525, "text": "tastes like college!", "updated": "2010-‐07-‐22 20:00:20" } link to comments link to beer { "id": "f1e62" } Saturday, October 6, 12

52 How to: look up comments from a beer •
SERIALIZED LOOP ﬁgure hrp://www.ibm.com/developerworks/webservices/library/ws-‐sdoarch/ beer = client.get(“beer:A_cold_one”); beer.comment_ids.each { |id| comments.push(client.get(id)); } • FAST MULTI-‐KEY LOOKUP beer = client.get(“beer:A_cold_one”); comments = client.multiGet(beer.comment_ids) • ASYNC VIEW QUERY comments = client.query(“myapp”,“by_comment_on”, {:key => “beer:A_cold_one”}); Saturday, October 6, 12

53 Emergent Schema JSON.org Github API Twiqer API "Capture the
user's intent" • The database can handle it • Your app controls the schema Saturday, October 6, 12

Audience par^cipa^on! *that means you 54 npm install twitterfight npm
start twitterfight Saturday, October 6, 12

@jchris hrp://www.couchbase.com/ Chris Anderson Thank You! 55 Saturday, October 6,
12

5 INCREMENTAL MAP-‐REDUCE FOR REALTIME ANALYTICS Saturday, October 6, 12

What do you mean “Incremental?” like: CREATE INDEX city ON
brewery city; 57 Saturday, October 6, 12

5 QUERY PATTERN: FIND BY ATTRIBUTE Saturday, October 6, 12

59 Find documents by a speciﬁc arribute • Lets ﬁnd
beers by brewery_id! Saturday, October 6, 12

60 The index deﬁni^on Saturday, October 6, 12

61 The result set: beers keyed by brewery_id Saturday, October
6, 12

6 QUERY PATTERN: BASIC AGGREGATIONS Saturday, October 6, 12

63 Use a built-‐in reduce func^on with a group query
• Lets ﬁnd average abv for each brewery! Saturday, October 6, 12

64 We are reducing doc.abv with _stats Saturday, October 6,
12

65 Group reduce (reduce by unique key) Saturday, October 6,
12

6 QUERY PATTERN: TIME-‐BASED ROLLUPS WITH KEY GROUPING Saturday, October
6, 12

group_level=3 -‐ daily results -‐ great for graphing 67 •
Daily, hourly, minute or second rollup all possible with the same index. • hrp://crate.im/posts/couchbase-‐views-‐reddit-‐data/ Saturday, October 6, 12

6 GEO INDEX & FULL TEXT INTEGRATION Saturday, October
6, 12

6 GeoCouch R-‐Tree Index • Op^mized for bulk loading of
large data sets • Simpliﬁed query model (bounding box, nearest neighbor) Saturday, October 6, 12

Elas^c Search Adapter 70 • Elas^c Search is good for
ad-‐hoc queries and faceted browsing • Our adapter is aware of changing Couchbase topology • Indexed by Elas^c Search aOer stored to disk in Couchbase Saturday, October 6, 12

@jchris hrp://www.couchbase.com/ Chris Anderson Thank You! 71 Saturday, October 6,
12

NoSQL Landscape: Speed, Scale, and JSON

NoSQL Landscape: Speed, Scale, and JSON

Transcript

More Decks by Chris Anderson

Other Decks in Programming

Featured