Couchbase Live Europe 2015: What’s New in Couchbase: Top 10 Capabilities that Eliminate Barriers to Availability, Performance & Scale

What’s New in CouchbaseTop 10 Capabilities in 3.0

Dipti BorkarSr. Director | WW Solution Engineering

Couchbase Server Overview

©2014 Couchbase Inc.

Couchbase provides a complete Data Management solution

3

High availability

cache

Key-value store

Document

database

Embedded database

Sync management

Multi-purpose capabilities support a broad range of apps and use cases

Enterprises often start with cache, then broaden usage to other apps and use cases


What makes Couchbase unique?

4

Performance & scalability

leaderSub millisecond latency with high throughput; memory-centric architecture

Multi-purpose

Simplified administrationEasy to deploy & manage; integrated Admin Console, single-click cluster expansion & rebalance

Cache, key value store, document database, and local/mobile database in single platform

Always-on availability

Data replication across nodes, clusters, and data centers

Enterprises choose Couchbase for several key advantages

24x365


Journey so far…

Couchbase Server v1.8High Scale Key/Value Database

Couchbase Server v2.1 Improved XDCRAdvanced Storage Engine

Couchbase Server v2.2Enhanced Security HA/DR

Couchbase Server v2.5Rack/Zone Awareness Connection Management

Couchbase Mobile 1.0Offline Data Availability

Auto Synchronization

Couchbase Server v3.0Mission Critical Scale & Perf

Simplified AdministrationImproved HA/DR & Security

2010

Couchbase Server 4.0With N1QLSQL for Documents

Couchbase Server v2.0Document ExtensionsNew XDCR


DEVELOPER EMPOWERMENT

MISSION CRITICAL SCALE

ULTRA HIGH AVAILABILITY

SECURE COMMUNICATION

SIMPLIFIED ADMINISTRATION

• Fully Integrated JSON Programmability

• Native Async and Reactive Programmability

• New Integration with Popular Frameworks

• N1QL Preview Support in SDKs

• Optimized Massive Databases with Tunable Memory

• Auto-Tuning Thread Pool for Extreme IO Throughput

• Faster Replication with DCP

• Improved Durability

• Faster view building

• Low Latency, In-Memory XDCR

• Delta Node Recovery

• Faster Rebalance

• Faster Node Restart with Efficient Warm-up

• New Incremental Backup option

• Encryption for Data Access

• Encryption for Administrative Access

• XDCR Pause and Resume

• Improved Resource Governance

• Graceful Failover

• Cluster-wide Diagnostics

What’s new in Couchbase Server v3.0

#0The Foundation


New DCP (Database Change Protocol) protocol for cluster-wide communication Eliminate Disk-IO from the replication access path High-Performance, Stream-based, In-memory Ordered based on global Sequence Numbers per partition Snapshot-able Restart-able

• Powers Node to Node Replication• Powers View • Powers XDCR

Database Change Protocol

This is non-trivial stuff!!!

9©2014 Couchbase Inc.

33 2

XDCR after Write

Managed Cache

Dis

k Q

ueue

Disk

Replication Queue

App Server

Couchbase Server Node

Doc 1

Doc 1

XDCR Queue

Doc 1

Doc 1

(New in 3.0) Memory-to-Memory Replication to remote cluster

Memory-to-Memory Replication to other node


33 2

View processing after write

Managed Cache

Dis

k Q

ueue

Disk

Replication Queue

App Server

Couchbase Server Node

Doc 1

Doc 1

To other node

View engine Doc 1

Doc 1

#1Tunable memory


Optimized Memory Usage with Metadata Ejection Policy Better optimization of memory for massive databases Enable efficient management of rarely accessed data set Cache only keys and data for the working set & eject all

historic data

Tunable Memory - Optimization for Massive Databases

©2014 Couchbase, Inc. ©2014 Couchbase Inc.

Each hash table item consists of {key, metadata, value} Metadata memory overhead is 40 bytes at least

A Key in the Hash Table

13

Key Metadata Blob pointer

Blob value

Expiration time

CAS identifier

Sequence number (DCP)

Revision number (XDCR)

Lock expiry (GetLocked API)

Flag, NRU, …

Hash Table Item

Pointer to next item


Application’s entire key space is maintained in the hash table

Highly cache-oriented architecture

Item pager ejects only an item’s value from the hash table

Value-Only Ejection

14

Key: “foo” Metadata Blob pointer

Blob valuePointer to next item

Hash Table ItemStorage Engine

Batch Reader

Get(“foo”)

read_value(“foo”)


Application’s entire key space doesn’t need to be loaded in cache

Reduce the memory overhead significantly in heavy DGM (Disk Greater than Memory) cases

Item pager ejects an item’s key and metadata along with its value

Full Metadata Ejection

15

Key: “foo” Metadata Blob pointer

Blob valuePointer to next item

Hash Table ItemStorage Engine

Batch Reader

Get(“foo”)

read_meta_value(“foo”)


Tunable Memory - Optimization for Massive Databases

100sx Reduction in Metadata Memory Consumption from 2.5 to 3.0

Note: The graph represents characteristics under data mutations. ~50M docs with value size ~0.5KB

v.3.0 – Large DB with Hot Working-set

3 GB Consumed for Metadata in RAM

2.5.1 or earlier – Real-time Latency

80 MB Consumed for Metadata in RAM

#2Recovery


Delta Node Recovery Shorter time to add nodes back into the cluster! Bring back a server incrementally into the cluster without

having to rebuild the full node.

Incremental Node Catch-up with Delta Node Recovery

100sx Reduction in Time to Re-Add Node from 2.5 to 3.0Note: The absolute performance improvement depend on data size and mutation count that needs to be caught up

#3Persistence and

I/O


Efficient Auto-Tuning Engine Detect and allocate threads based on HW resources Pool threads for best resource utilization across buckets Improved latency across the board:

Faster ReadsFaster RebalanceFaster Node reactivation Faster Durability with Writes & PersistTo (more

details on this on later)…and many more

Auto Tuning Thread Pool


Inside data manager

21

Append-only B-Tree Storage Engine

Engine APIs(get, set, del, add, append, DCP,

…)

PartitionHash Table

(active)

PartitionHash Table

(replica)

PartitionHash Table

(active)…

Checkpoints

Checkpoints

Checkpoints

ReaderThreads

Non-IO Thread

s

DataReplicato

r

I/O Completion

Notifier

Aux-IOThreads

FlushersData

Backfill

User Configured Replica Count = 1

Batch Readers

WriterThreads

…

Item Pager

Expiry Pager

Checkpoint Manager

Shared Thread Pool


Efficient warm-up Parallelization on cache warm-up Further warm-up time improvements under metadata ejection

Up to 60% Faster Warm-up between 3.0 and 2.5.1 Up to 8x Faster Warm-up with Full Ejection in 3.0

*Note: Absolute latency depend on disk IO latency and bandwidth

Faster Node Re-Activation


Priority Setting on Buckets Ability to assign high and low priority to buckets High priority buckets get higher IO priority and more

throughput Low priority buckets can still be processed without

starvation

Improved Resource Governance with Bucket Priority

#4Replication


Faster Durability with Replication

Improved latency on ReplicateTo DCP Improves replication speed and durability Better protects your data!

Up to 150x Improvement on ReplicateTo latency from 2.5 to 3.0

*Note: Absolute latency depend on network bandwidth and latency

*95th percentile ReplicateTo=1 latency (ms), 1 bucket x 200M x 1KB, 250 mutations/sec/node

Faster Replication within the cluster & across clusters

0

200

400

600

2.5.1 3.0

95 Percentile Repli-cateTo=1 Latency*

Tim

e (

ms)


Improved Protections for Your Data

Drastically Lower XDCR Latency In memory change detection with DCP

No longer require persisting to disk

Up to 4x better on XDCR latency between clusters between 3.0 & 2.5.1

*Note: Absolute latency depend on WAN latency and bandwidth

Lower Latency across clusters

0

100

200

300

400

500

600

2.5.1 3.0

90th Percentile. XDCR Latency

Tim

e (

ms)

90th percentile replication lag (ms), 5 -> 5 UniDir, 2 buckets x 500M x 1KB, 10K SETs/sec, LAN

#5Indexing data


More up-to-date views Mutation show up at the

indexes faster with the new DCP protocol

No longer require persisting to disk for change detection

Improved consistency with views (stale=false behavior)

Ports to C in view engine

Very fast index building

05000

10000150002000025000300003500040000

2.5.1 3.0

Indexing Under Load

Tim

e (

ms)

Up to 50x Faster Indexing from 2.5 to 3.0*Note: Absolute latency depend on disk IO speed

*95th percentile Indexing latency (ms), 1 bucket x 20M x 2KB, non-DGM, 1 view, 250 mutations/sec/node, 400 queries/sec

#6Backup


Minimize time and resources during backups

Efficient Recovery with Incremental Backup & Restore

• Back up only the data updated since the last backup

• Differential Backups• Cumulative Backups

#7XDCR


XDCR Pause and Resume Want to suspend replication? No longer need to drop

replications Ability to pause XDCR traffic anytime Resume and incrementally catch-up from where you left

off

Efficient Maintenance with XDCR

#8Connection

management


Connection Management

Optimized Connection Management

Cluster map / topology information transmitted via 11210 port rather than 8091

Much more scalable and better performance during topology changes


Connection Management

Increased connection limits

Data port : 11210 max_conns_on_port_11210: 30000Internal Cluster communication : 11209 max_conns_on_port_11209: 5000

#9Maintenance


New option to gracefully failover a node. Failover with a graceful handoff May take longer to failover but prevents potential for any

data loss

Node Maintenance

#10SDKs


Componentized libraries to simplify use cases• Couchbase JVM Core - Reusable Foundation

• Handles core IO, topology changes. Reactive and asynchronous.• Java Client Library

• Native JSON handling built-in• Provides synchronous and asynchronous interface• Preview of N1QL language integrated query

What’s New in SDK 2.0


Integrated native JSON data type programmability

• Simplified Development • Native JSON handling built-in• Preview of N1QL language integrated query

• Performance & Scale• Improved connection management

• Native .Net: • Simplified Configuration Management with Web.config & App.config



Node.js, PHP• Native JSON handling built-in• Core IO enhancements through updated C libcouchbase• Better connection management

Python, Ruby• Core IO enhancements through updated C libcouchbase• Better connection management


What’s ahead

Couchbase Roadmap2015 – Couchbase Server 4.0

POWERFUL QUERYPERFORMANCE &

SCALEHIGH

AVAILABILITYSECURITY

COMPLIANCEDEVELOPER

EMPOWERMENT

• N1QL: SQL for Documents

• Secondary indexing for interactive queries

• Improved Spatial Queries

• Multi Dimensional Scaling

• Scalability Enhancements for Massive Database Deployments

• Improve Performance for Full Eviction – Bloom Filters

• Filtering Replication for Efficient Geo Data Distribution with XDCR

• Auditing Framework and Admin Auditing

• External Account Management System Integration for the Enterprise (LDAP)

• Enhanced Consistency Options

• Legacy Data interfaces ODBC/JDBC


Introduction to N1QL – SQL for Documents Next generation, NoSQL query language

SQL-like : SELECT * FROM WHERE/LIKE/JOIN/GROUP/etc, CREATE INDEX Extended for JSON to support nested and hierarchical data structures Support for views and newly-developed secondary indexes Query (DQL), Manipulation (DML), Description (DDL) ODBC/JDBC drivers in development

Built into Couchbase Server: Single installation package Multi-threaded, stateless query and indexing components Leverages high-performance, high-scale Couchbase buckets

Coming in 2015, preview at query.couchbase.com


N1QL Architecture

Single node installation, services defined dynamically

Query service access Index and Data to formulate response

All queries and direct access is topology aware and dynamically scalable

Q & AThank you.

[email protected]@dborkar

Couchbase Live Europe 2015: What’s New in Couchbase: Top 10 Capabilities that Eliminate Barriers to Availability, Performance & Scale

Software

memory replication

couchbase unique

couchbase server v3

couchbase server overview

availability data replication

remote cluster memory

metadata memory overhead

optimized memory usage