In-Memory Computing - Distributed Systems - Devoxx UK 2015

www.hazelcast.com

/** * In-Memory Computing * Distributed Systems */

www.hazelcast.com@noctarius2k

My Current View

www.hazelcast.com

DisclaimerThis presentation may include certain forward-looking statements and projections provided by the Company. Any such statements and projections reflect various

estimates and assumptions by the Company concerning anticipated results. These statements involve risks, uncertainties and assumptions and are based on the

current estimates and assumptions of the management of the Company as of the date of this presentation and are subject to uncertainty and changes. No

representations or warranties are made by the Company as to the accuracy of any such statements or projections. Whether or not any such forward-looking statements or projections are in fact achieved will depend upon future events some of which are not within the control of the Company. Accordingly, actual results may vary from the projected results and such variations may be material. Statements contained herein describing documents and agreements are summaries only and such summaries

are qualified in their entirety by reference to such documents and agreements.

www.hazelcast.com

Who’s that dude?

• Chris(toph) Engelbert • Twitter fanatic: @noctarius2k • Weird 9+ Java-Years • Performance, GC, Fairytales • Apache Committer • Gaming, Travel Management, …

www.hazelcast.com

Who’s that dude?

• Chris(toph) Engelbert • Twitter fanatic: @noctarius2k • Weird 9+ Java-Years • Performance, GC, Fairytales • Apache Committer • Gaming, Travel Management, …

The www.hazelcast.ninja


DistributedComputing…


ToasterServer 1

TV-Server 1Power

CoffeeMakerServer 1

Kettle Server 1

StoveServer 1

Power GridServer 1, 2

DistributedComputing…


… but what meansDistributed Computing?OOO

( )

www.hazelcast.com


‘In-Memory’ -Don’t you mean ‘Wasting Memory’?

L1 Cache Reference L2 Cache Reference Main Memory Reference 1K send over 1 Gbps Network 4K read from SSD Seek on Spinning Disk Send Packet CA->AMS->CA

0.5 ns 7 ns

100 ns 10,000 ns

150,000 ns 10,000,000 ns

150,000,000 ns

0.01 ms 0.15 ms

10 ms 150 ms


Evolution of:- Memory-Price- Memory-Size

www.hazelcast.com

Data Distribution(Partitioning / Sharding)


Data Distribution

String code = “devoxx.co.uk”

Map<String, TShirtDelivery> shirts = Maps.create()

shirts.put(code, new TShirtDelivery(“Your Address”))


Data Distribution

String code = “devoxx.co.uk”

int hashcode = hash(code)

partitionId = hashcode % partitionCount

cluster.send(partitionId, code, delivery)


Data Distribution

int hashcode = hash(code)

partitionId = hashcode % partitionCount

cluster.send(partitionId, code, delivery)

Must be a consistent hash function

Partition count must be constant

Can be calculated, smart routing possible


Demo Time


That was easy!But don’t tell anyone!

www.hazelcast.com

Parallelize Processing



int[] values = [0..100000].map(randomInt())

int[][] blocks = split(values, 10000)

int[] subresult = blocks.foreach(_.sum())

int result = subresult.sum()



int[][] blocks = split(values, 10000)

int[] subresult = blocks.foreach(_.sum())

int result = subresult.sum()

No mutation, only new values

Independently computable

Might split further down (fork-join, work-stealing)


Demo Time2


I told you:Easy Peasy!


Caching!


Collect, store …


…access fast when necessary!…access fast when necessary!


http://gadizmo.com/roll-your-own-with-the-ham-dogger.php

Roll Your Own!Evolution Of Caching

http://gadizmo.com/roll-your-own-with-the-ham-dogger.php


Caches Are Simple!

public class Cache<K, V> { private final Map<K, V> cache = new CHM<>();

public void put(K key, V value) { cache.put(key, value); }

public V get(K key) { return cache.get(key); }

}


Even With TTL!


And Auto-Cleanup


What Is Wrong?


What Is Wrong?

O(n)5 mio

Entries



Evolution Of Caching


Proprietary Cache!



Proprietary Cache!


Open Source


Proprietary Cache!


Open Source

Commercial


Proprietary Cache!


Open Source

Commercial

Stupidly Expensive


http://technoodling.net/mili-power-crystal-battery-pack-review/

Terracotta /EHcache

OracleCoherence

Infinispan

JavaCollections

API

Hazelcast

JCache

http://technoodling.net/mili-power-crystal-battery-pack-review/


Success!Feels like


Want to know moreabout #JCache?

Gimme CachingThe Distributed JCache (JSR107) Way

17:00 - 17:50 - Room Exec Centre

David Brimley


Quick Shameless Plug :)Hazelcast!


Hazelcast in 5 Facts• Java Collections API• Java Concurrency API

• Transparent Data Distribution• Drop-In Replacement

• Disruptively Simple(c) Rod Johnson (@springrod)


Time to be Happy!


Time to be Happy!

Apache License 2


Time to be Happy!

www.hazelcast.org


You can do it!

www.hazelcast.com

Thank You!Any Questions?

@noctarius2k http://www.sourceprojects.org

http://github.com/noctarius

@hazelcast http://www.hazelcast.com http://www.hazelcast.org

http://github.com/hazelcast

In-Memory Computing - Distributed Systems - Devoxx UK 2015

Technology

memory computing

data distribution int

hashcode partitionid

memory dont

forwardlooking statements

distributed computing

stove server

kettle server