Антон Бойко "Разделяй и властвуй — набор практик для построения масштабируемого приложения в облаке"

Спикер:

Тема презентации:

Антон Бойко, Microsoft Azure MVP

Разделяй и властвуй — набор практик

для построения масштабируемого приложения в облаке

Who am I?

Developer• .NET – 10+ years• Node.JS and JavaScript – 3+ years

Cloud engineer• Microsoft Azure – 4+ years• Amazon Web Services – 3+ years

Community leader• Microsoft Azure MVP – 2+ years• Ukrainian Microsoft Azure Community Founder

Consultant• 30+ cloud based projects (from 5 to 5M users) including startups, migration, etc.• 20+ cloud based trainings, workshops, webinars, etc.

Agenda

• Scalability? It is and it isn`t.• Ways of scaling.• Divide and conquer. Divide what and conquer who?• Go into more details with scalability patterns.

Scalability

Scalability is…

• Scalability is the ability of a system, network, or process to handle a growing amount of work in a capable manner or its ability to be enlarged to accommodate that growth.

Scalability story #1 – Gnutella

• When a user wants to do a search, a client sends a request to each actively connected node. In version 0.4 of the protocol, the number of actively connected nodes for a client was quite small (around 5), therefore each node forwarded the request to all its actively connected nodes, and they in turn forwarded the request, and so on, until the packet reached a predetermined number of hops from the sender.

Scalability story #2 – Domain Name System

.

.com.dataart .blog

.google

.ua .in.testing

.teamspark.azureday

.gwab

Scalability is not…

• Performance tuning

Tuning Scaling

Ways of scaling

Vertical scaling

1 CPU4 GB RAM

4 CPU16 GB RAM

X4

Vertical scaling

Pros• Simple and straightforward• Better resources’ utilization

Cons• Usually requires shutdown

(downtime) to scale

Horizontal scaling

1 CPU4 GB RAM

1 CPU4 GB RAM

1 CPU4 GB RAM

1 CPU4 GB RAM

1 CPU4 GB RAM

X4

Horizontal scaling

Pros• Scale without downtime• Improved availability out of the

box

Cons• Your app needs to support

horizontal scaling• Can induce inefficient resources’

utilization

There is no silver bullet

Divide and conquer

Divide application

• Split your app into independent modules which can perform some small and dedicated tasks. For example:• Client• API• Application• Data access

• Do not try to do all tasks at once, you need to eat your elephant piece by piece.• It’s much easier and more transparent to work within simple module

scope than the system scope itself.

Conquer workload

• If you know your enemies and yourself, you will not be imperiled in a hundred battles - Sun Tzu, The Art of War.• Determine whom you are fighting against.• Know your strengths and know your weaknesses.

Anyway… Always use your head!

Divide and conquer story #1 – web site content analyzer

Downloader queue Downloader

Content storage

Analyzer queue

Results storage

Analyzer

APIClient App

1..N

1..N

1..N

Divide and conquer story #2 – Map Reduce

Sheldon: I made tea. Leonard: I don’t want tea. Sheldon: I didn’t make tea for you. This is my tea. Leonard: Then why are you telling me? Sheldon: It’s a conversation starter. Leonard: That’s a lousy conversation starter. Sheldon: Oh, is it? We’re conversing. Checkmate.

HDFS Map Reduce Result

Raw data(line from txt file)

Key and raw data array Key-value pairs

“I made tea.” =>“tea” : “I made tea.”

“tea” : [“…”, “…”, “…”] =>“tea” : “3”

“tea” : “3”“you” : “1”

Scalability patterns

Static content hosting

Web Site

Web SiteWeb Site

Static content hosting CDN

Static content hosting

Pros• Simple and straightforward way

to decrease web server load• Very easy to implement• Bring static content closer to end

users

Cons• Need to wait for TTL to expire on

caching node or need to update caching node explicitly• Can supply content from

different IP address• Can be harder to deploy new

version because of system inconsistency

Worker

Queue-based load leveling

Web Site

Worker

Queue-based load leveling

Web Site

Queue

Worker

Queue-based load leveling

Pros• Easy way to scale independent

atomic tasks

Cons• Can take some time to rearrange

system communication mechanism• Queue is a one-way

communication channel• Queue can increase response

latency

Command and query responsibility segregation

CRUD

Command and query responsibility segregation

Write

Read

Sync

Command and query responsibility segregation

Pros• Provides an ability to manage

writes and reads load separately• Reads can be easily scaled

horizontally

Cons• Can take a lot of time to migrate

from regular CRUD based code• Can easily become

unmanageable and cumbersome

Sharding

CharleyJudithJake

Sharding

Charley Judith Jake

CharleyJudithJake

Sharding

Pros• Can scale data storage almost to

infinity• Can easily manage data load for

different shards

Cons• Impossible to use native queries

to query data among several shards (unless your engine is able to do that)• Cross-shard queries can require

a lot of additional coding• Will introduce additional

redundancy

And again… Always use your head!

Q & AAnton Boyko

boyko.ant@live.com

facebook.com/groups/azure.ua

it-community.in.ua