24
In a constantly changing industry you need a forward-thinking technology partner Cassandra and Hybrid-Cloud: Mache Neil Avery, CTO Email: [email protected] Twitter: @avery_neil

Cassandra and Hybrid Cloud - Introducing Mache

Embed Size (px)

Citation preview

In a constantly changing industry you needa forward-thinking technology partner

Cassandra and Hybrid-Cloud: Mache

Neil Avery, CTOEmail: [email protected]: @avery_neil

2

Cassandra and the move the Hybrid cloud - [email protected]

Cloud is seen as a solution within financial services

Key Drivers: - Regulations, Cost, Data-Centre, Capacity- Running a Data-Centre is hard

Grid is the ‘Quick Win’

Projected AWS revenue in 2016: $1.8b

3

Cassandra and the move the Hybrid cloud - [email protected]

Compute Grids seem like a quick-win.

- Large scale, DC footprint

- Service most of the banks function

- Consume majority of DC estate

Scale: 10k-100k cores per app

4

Cassandra and the move the Hybrid cloud - [email protected]

Grid is a ‘quick win’ but…

- Cloud isn’t – public cloud sucks

- Security, compliance, outsourcing

- Hybrid Cloud helps…

5

Cassandra and the move the Hybrid cloud - [email protected]

What is Hybrid Cloud…?

6

Cassandra and the move the Hybrid cloud - [email protected]

Source: http://www.csc.com/cloud/insights/105069-hybrid_cloud

7

Cassandra and the move the Hybrid cloud - [email protected]

Hybrid Cloud Challenges affecting Grid

The key thing here is:

The network is different – leading to :

security, operations, integration challenges

The grid-application architecture will also be different

8

Cassandra and the move the Hybrid cloud - [email protected]

What do compute grids look like?

Typical numbers for a standard grid-app

- 30k cores/engine

- 50m tasks

- 7Mb per task

- 184GB peak data

- 10,500 GB daily

Numbers for a ‘data & compute’

- 18k cores/engines

- 7.5+GB per engine

- Peak: 144k GB – we can

share – 375 servers - or

385GB per server

9

Cassandra and the move the Hybrid cloud - [email protected]

Adopting grid + hybrid cloud = Data locality challenge

- 2 Cloud regions: 2 x (16 datacentres, 80k servers)

- 2 Datacentres (1-region)

- Cloud Exchange

- 3 locations for data to live

- Data has a life-cycle

- Thin-pipes == data-gravity - : 144k GB ?

10

Cassandra and the move the Hybrid cloud - [email protected]

The solution: Cassandra ‘DataCentres’

Mr Thin Pipes versus Cassandra

11

Cassandra and the move the Hybrid cloud - [email protected]

Mapping Cassandra onto Cloud regions

Cloud Region A

Source: http://docs.datastax.com/en/cassandra/2.2/cassandra/dml/dmlClientRequestsRead.html

Network Topology

Cloud Region B

Cloud Exchange

On-Prem

12

Cassandra and the move the Hybrid cloud - [email protected]

We need to solve the problem of

- Massive data

- Massive Compute (read-mostly)

- Thin pipes

- Data Pools

Be nice to the network

- NetworkTopologyStrategy

- LOCAL_QUORM

- REPLICATION_FACTOR

13

Cassandra and the move the Hybrid cloud - [email protected]

Cassandra allows us to shape the data layer

… but we have long-lived, high-read frequency, massive-data

‘16MB x 200 x 10k – read-mostly data’

Answer: Nearside cache

14

Cassandra and the move the Hybrid cloud - [email protected]

Why not use an IMDB? (Coherence, Gemfire, Gigaspaces)

- 16MB x 200 items per task

- 10k clients

- Leads to network DoS

IMDB’s only read from a primary-object- ‘Read’s don’t scale’

15

Cassandra and the move the Hybrid cloud - [email protected]

Introducing Mache….

Nearside-Cache = ‘mash-up of Open source’ htthttps://github.com/Excelian/Machetthttps://github.com/Excelian/MacheDistributed nearside-cache with eventing/invalidation

- Cassandra

- Google Guava Cache

- Spring Data

- Kafka, ActiveMq, Rabbit

- Yes – it’s java only

16

Cassandra and the move the Hybrid cloud - [email protected]

- Avoid heavy write loads:can cause event storms

- Read-mostly – data consistency- Infrequently changing data- Read with some write – Kafka

Mache: Use case

17

Cassandra and the move the Hybrid cloud - [email protected]

Mache: Performance environment

- Cassandra 2.x (4 nodes)

- Docker, Kubernetes, Fabric8,

- SDN: Project Calico

18

Cassandra and the move the Hybrid cloud - [email protected]

Cassandra: PerformanceCaveat: office network

19

Cassandra and the move the Hybrid cloud - [email protected]

Mache: PerformanceCaveat: office network

20

Cassandra and the move the Hybrid cloud - [email protected]

Mache: Update Frequency v LatencyCassandra, Mache & Kafka

21

Cassandra and the move the Hybrid cloud - [email protected]

Mache: Code example1) Cassandra Data

2) Creating the cache

3) Using the cache

22

Cassandra and the move the Hybrid cloud - [email protected]

Next steps

- Rest/JSON service

- JMX Support

- Nearside Overflow to disk

- Selective event to enable continuous query

- PoC’s

- Eventing API (listener)

Mache: available now!https://github.com/Excelian/Mache

23

Cassandra and the move the Hybrid cloud - [email protected]

Key takeaways:

• Cloud has found its home within FS

• Hybrid cloud represents new challenges

• Cassandra solves ‘data-gravity’ problem

• Nearside caching offloads high frequency reads from the storage

layer

https://github.com/Excelian/Mache

Cape Town Office1st Floor, Corporate Place

13 Mispel Road, BellvilleCape Town, 7530

South Africa+27(0) 21 944 9900

Frankfurt OfficeBrockenheimer Landstrasse 17/19

60325 FrankfurtGermany

+49 (0) 69 710455 183

London Office44 Featherstone Street

London, EC1Y 8RNUnited Kingdom

+44 (0)20 7336 9595

New York OfficeSuite 503, 100 Wall Street

New York, NY 10005

United States

Poland Officeul. Plac Konstytucji 3 Maja, 3, Silver

Tower

50-048 Wrocław

Poland

Sydney Office1 York Street

Level 2Sydney, NSW 2000

Australia+61 (2) 9191 7810

Toronto Office100 King Street West, Suite 5600

Toronto, Ontario M5X 1C9Canada

+1 (203) 295 5240

www.excelian.com