sac-853t: inside windows azure: the cloud operating...

= Managed for You Standalone

Servers

IaaS PaaS SaaS

Applications

Runtimes

Database

Operating System

Virtualization

Server

Storage

Networking

Windows Azure

Efficiency

Control+Cost

2) Choose image,

then create and

configure VM(s)

for application

1) Choose image, then

create VM for DBMS

and configure DBMS

Library

VM Images

Developer

ApplicationData

Load

Balancer

5)

Configure

load

balancer

6) Manage VMs and

DBMS (e.g., deploying

new OS images in VMs)

3) Provision

database, then

create tables and

add data

4) Install

application

Developer

ApplicationData

Load

Balancer

2) Deploy

application

1) Provision

database, then

create tables and

add data

Windows Azure

Networking

“Red Dog” Front End (RDFE)

DLA Architecture (Old) Quantum10 Architecture (New)

TOR TOR TOR TOR

Spine Spine Spine

…

…

DCR DCR

LBLB

Spine

DC Routers

LB LB

40 Nodes

TOR

L

B

L

B

AGG

Digi

APC

L

B

L

B

AGG

L

B

L

B

AGG

L

B

L

B

AGG

L

B

L

B

AGG

L

B

L

B

AGG

20Racks

DC Router

Access

Routers

Aggregation +

LB

40 Nodes

TOR

Digi

APC

40 Nodes

TOR

Digi

APC

40 Nodes

TOR

Digi

APC

40 Nodes

TOR

Digi

APC

40 Nodes

TOR

Digi

APC

40 Nodes

TOR

Digi

APC

40 Nodes

TOR

Digi

APC

40 Nodes

TOR

Digi

APC

40 Nodes

TOR

Digi

APC

40 Nodes

TOR

Digi

APC

40 Nodes

TOR

Digi

APC

40 Nodes

TOR

Digi

APC

40 Nodes

TOR

Digi

APC

40 Nodes

TOR

Digi

APC

……

20Racks 20Racks 20Racks

…… … …

Server

WordSQL

Server

Datacenter

Exchange

OnlineSQL Azure

TOR TOR TOR TOR TOR

AGGLB LB

Fabric Controller

Role

Images

Role

Images

Role

Images

Role

Images

Image Repository

Maintenance OSParent

OS

Node

PXE

ServerMaintenance OS

Windows Azure

OS

Windows

Azure

OS

FC Host

Agent

Windows Azure Hypervisor

Windows

Deployment

Server

Role BWorker Role

Count: 2Update Domains: 2

Size: Medium

LoadBalancer

www.mycloudapp.net

www.mycloudapp.net

Fabric Controller (Primary)

FC Host Agent

Host Partition

Guest

Partition

Guest

Agent

Guest

Partition

Guest

Agent

Guest

Partition

Guest

Agent

Guest

Partition

Guest

Agent

Physical Node

Fabric Controller (Replica) Fabric Controller (Replica)…

Role Instance Role Instance Role Instance Role Instance

Trust boundary

Image Repository (OS VHDs,

role ZIP files)

• Blobs

• Tables

• Queues

BlobContainerAccount

sally

pictures

IMG001.JPG

IMG002.JPG

movies MOV1.AVI

EntityTableAccount

sally

users

Name =…

Email = …

Name =…

Email = …

photo index

Photo ID =…

Date =…

Photo ID =…

Date =…

MessageQueueAccount

sally

thumbnail jobs

128x128, http://…

256x256, http://…

photo processing jobs

http://…

http://…

Account

Container Blobs

Table Entities

Queue Messages

http://<account>.blob.core.windows.net/<container>

http://<account>.table.core.windows.net/<table>

http://<account>.queue.core.windows.net/<queue>

Front-

End-1

Front-

End-2

Middle

Tier-2

Middle

Tier-1Middle

Tier-3

Front-

End-1

Middle

Tier-1

Front-

End-2

Middle

Tier-2Middle

Tier-3

25 min

Guest

Agent

Connect

Timeout

Guest Agent

Heartbeat

5s

Role

Instance

Launch

Indefinite

Role

Instance

Start

Role

Instance

Ready

(for updates only)

15 min

Role Instance

Heartbeat

15s

Guest Agent

Heartbeat Timeout

10 min

Role Instance

“Unresponsive” Timeout

30s

Load Balancer

Heartbeat

15s

Load Balancer

Timeout

30s

Guest Agent

Role Instance

Front-End

Front-End

Windows

Azure

Storage,

SQL Azure

Load Balancer

Middle-

Tier

Queue

Using queues

Web Role

ASP.NET, WCF, etc.

Worker Role

main(){ … }

1) Receive work

2) Put work in queue

3) Get work from queue

4) Do work

To scale, add more of either

Queues are the application glue• Queues decouple different parts of application, making it

easier to scale app parts independently;

• Flexible resource allocation, different priority queues and separation of backend servers to process different queues.

• Queues mask faults in worker roles.

BLAST (Basic Local Alignment Search Tool)

• Most important software in bioinformatics

• Identify similarity between bio-sequences

Computationally intensive

• Large number of pairwise alignment operations

• A BLAST running can take 700 ~ 1000 CPU hours

• Sequence databases growing exponentially

• GenBank doubles in size every 15 months.

It is easy to parallelize BLAST

• Segment the input

• Segment processing (querying) is pleasingly parallel

• Segment the database

• Needs special result reduction processing

Large volume data

• A normal Blast database can be as large as 10GB

• 100 nodes means the peak storage bandwidth could reach to 1TB

• The output of BLAST is usually 10-100x larger than the input

• Parallel BLAST engine on Azure

• Query-segmentation data-parallel pattern

• split the input sequences

• query partitions in parallel

• merge results together when done

• Follows the general PaaS application model

• Web Role + Queue + Worker

Wei Lu, Jared Jackson, and Roger Barga, AzureBlast: A Case Study of Developing Science Applications on the Cloud, in Proceedings of the 1st Workshop on

Scientific Cloud Computing (Science Cloud 2010), Association for Computing Machinery, Inc., 21 June 2010

http://research.microsoft.com/apps/pubs/default.aspx?id=122754

A simple Split/Join pattern

Leverage multi-core of one instance • argument “–a” of NCBI-BLAST

• 1,2,4,8 for small, middle, large, and extra large instance size

Task granularity • Large partition load imbalance

• Small partition unnecessary overheads• NCBI-BLAST overhead

• Data transferring overhead.

Best Practice: test runs to profiling and set size to mitigate the overhead

Value of visibilityTimeout for each BLAST task, • Essentially an estimate of the task run time.

• too small repeated computation;

• too large unnecessary long period of waiting time in case of the instance failure.

BLAST task

Splitting task

BLAST task

BLAST task

BLAST task

…

Merging Task

Task size vs. Performance• Benefit of the warm cache effect

• 100 sequences per partition is the best choice

Instance size vs. Performance• Super-linear speedup with larger size

worker instances

• Primarily due to the memory capability.

Task Size/Instance Size vs. Cost• Extra-large instance generated the best

and the most economical throughput

• Fully utilize the resource

Web Portal

Web

Service

Job registration

Job Scheduler

Worker

Worker

Worker

Global

dispatch

queue

Web Role

Azure Table

Job Management Role

Azure Blob

Database

updating Role

…

Scaling Engine

Blast databases,

temporary data, etc.)

Job Registry

NCBI databases

BLAST task

Splitting task

BLAST task

BLAST task

BLAST task

…

Merging Task

Discovering Homologs • Discover the interrelationships of known protein sequences

“All against All” query• The database is also the input query

• The protein database is large (4.2 GB size)• Totally 9,865,668 sequences to be queried

• Theoretically, 100 billion sequence comparisons

Performance estimation• Based on the sampling-running on one extra-large Azure instance

• Would require 3,216,731 minutes (6.1 years) on one desktop

One of largest BLAST jobs• This scale of experiments usually are infeasible to most scientists

• Allocated a total of ~4000 instances • 475 extra-large VMs (8 cores per VM), four datacenters, US (2), Western and North Europe

• 8 deployments of AzureBLAST• Each deployment has its own co-located storage service

• Divide 10 million sequences into multiple segments• Each will be submitted to one deployment as one job for execution

• Each segment consists of smaller partitions

• When load imbalances, redistribute the load

50

62

6262

6262

5062

3/31/2010 6:14 RD00155D3611B0 Executing the task 251523...

3/31/2010 6:25 RD00155D3611B0 Execution of task 251523 is done, it took 10.9mins


3/31/2010 6:44 RD00155D3611B0 Execution of task 251553 is done, it took 19.3mins


3/31/2010 7:02 RD00155D3611B0 Execution of task 251600 is done, it took 17.27 mins



3/31/2010 11:12 RD00155D3611B0 Execution of task 251895 is done, it took 82 mins

North Europe Data Center, totally 34,256 tasks processed

All 62 compute nodes lost tasks and

then came back in a group. This is

an Update domain

~30 mins

~ 6 nodes in one group

35 Nodes experience blob

writing failure at same time

West Europe Datacenter; 30,976 tasks are completed, and job was killed

A reasonable guess: the

Fault Domain is working

IaaS PaaS Relational NoSQL BlobsIaaS

Amazon

Microsoft

VMware

OpenStack

Computing StoragePublicPrivate

Key

Cloud Platform

Software

Cloud Platform

Service

Microsoft Private

Cloud

(for hosters)

IaaS PaaS Relational NoSQL

Computing Storage

BlobsIaaS

PublicPrivate

Amazon

Microsoft

VMware

OpenStack

Key

Cloud Platform

Software

Cloud Platform

Service

Microsoft Private

Cloud

Core cloud technologies

implemented in System

Center 2012

vCloud


Computing Storage

BlobsIaaS

PublicPrivate

Amazon

Microsoft

VMwarevCloud

DataCenter

(for hosters)

OpenStack

Key

Cloud Platform

Software

Cloud Platform

Service

Microsoft Private

Cloud

Core cloud technologies

implemented in vSphere and

vCloud Director

Microsoft Private

Cloud

(for hosters)

Web/Worker Roles SQL

AzureTables Blobs


Computing Storage

BlobsIaaS

PublicPrivate

Amazon

Microsoft

VMware vCloud

OpenStack

Key

Cloud Platform

Software

Cloud Platform

Service

Microsoft Private

CloudMicrosoft Private

Cloud Persistent VM

Roles

vCloud

DataCenter

(for hosters)

Cloud Foundry Cloud Foundry Cloud Foundry



AzureTables

Computing Storage

Blobs

Blobs

IaaS

PublicPrivate

Amazon

Microsoft

VMware vCloud

OpenStack

Key

Cloud Platform

Software

Cloud Platform

Service

Microsoft Private


Cloud Persistent VM

Roles

vCloud

DataCenter

(for hosters)

Elastic Compute

Cloud (EC2)

Relational

Database Service

(RDS)

SimpleDB Simple Storage

Service (S3)

Elastic

Beanstalk DynamoDB



AzureTables

Computing Storage

Blobs

Blobs

IaaS

PublicPrivate

Amazon

Microsoft

VMware vCloud Cloud Foundry Cloud Foundry

OpenStack

Key

Cloud Platform

Software

Cloud Platform

Service

Microsoft Private


Cloud Persistent VM

Roles

Cloud Foundry

vCloud

DataCenter

(for hosters)

IaaS

Elastic Compute

Cloud (EC2)

PaaS Relational NoSQL


AzureTables

Relational

Database Service

(RDS)

SimpleDB

Computing Storage

Blobs

Simple Storage

Service (S3)

Blobs

Elastic

BeanstalkEucalyptus

IaaS

PublicPrivate

Amazon

Microsoft


OpenStack

Key

Cloud Platform

Software

Cloud Platform

Service

Microsoft Private


Cloud Persistent VM

Roles

DynamoDB

Cloud Foundry

vCloud

DataCenter

(for hosters)

OpenStack

Compute

OpenStack

Compute

(for hosters)

OpenStack

Object Storage

(for hosters)

IaaS

Elastic Compute

Cloud (EC2)

PaaS Relational NoSQL


AzureTables

Relational

Database Service

(RDS)

SimpleDB

Computing Storage

Blobs

Simple Storage

Service (S3)

Blobs

Elastic

Beanstalk

IaaS

PublicPrivate

Amazon

Microsoft


OpenStack

Key

Cloud Platform

Software

Cloud Platform

Service

Microsoft Private


Cloud Persistent VM

Roles

DynamoDB

Cloud Foundry

Eucalyptus

vCloud

DataCenter

(for hosters)

Interesting to people

building applications:

Developers

Interesting to people

running applications:

Operations

IaaS PaaS

Running Existing

Web Apps/Sites

Running Standard

Packaged Apps

Running a Standard

DBMS

High Performance

Computing and Big Data

VMs for a

Dev/Test Lab

Running New

Cloud-Native Apps

Disaster Recovery

Virtual Data Center

(VMs for On-Demand Use)

Developer Operations

Best Suited For

Reliability

Provided By

Main Benefits

Typical Buyer

Examples

Typical

Management Tools

A much bigger

market today

Leaders in Gartner

Magic Quadrant for

Public Cloud IaaS

IaaS TypeHypervisorOffering

CSC VMware OperationsCloudCompute

DeveloperHyper-VMicrosoftWindows Azure Persistent

VM Role

Amazon Xen DeveloperElastic Compute Cloud

(EC2)

HP KVM DeveloperCloud Compute

Rackspace Xen DeveloperCloud Servers

IBM KVM DeveloperSmartCloud

Enterprise

Terremark VMwareOperations,

Developer

Enterprise Cloud, vCloud

Express

Savvis VMware OperationsSymphony VPDC

Bluelock VMware OperationsBluelock Virtual

Datacenters

GoGrid Xen DeveloperCloud Servers

Storage

Languages/ Frameworks

Offering

HerokuRuby/Rails, JavaScript/ Node.js,

Java, …

Relational (MySQL, Postgres, …),

NoSQL (Redis, …)Heroku

Heroku runs on EC2 and is

owned by Salesforce

Amazon Java/ServletsRelational (RDS),

NoSQL (SimpleDB, DynamoDB), …Elastic Beanstalk

Beanstalk is a simple extension to

EC2

Google Java, Python, GoRelational (CloudSQL),

NoSQL (Datastore), BlobsApp Engine

App Engine has undergone many

recent changes

Salesforce Apex/AppForce FrameworkNoSQL

(Database.com)AppForce

Pricing is per user, not based on

resources used

Oracle Java/Java EE (WebLogic) Relational (Oracle DBMS)Oracle Public Cloud Announced October 2011

Relational (SQL Azure),

NoSQL (Tables), BlobsC# and VB/.NET, PHP,

JavaScript/Node.js, …Microsoft

Windows Azure

Web/Worker Roles

Designed to be a fully PaaS

platform

Engine YardRuby/Rails,

PHPRelational (MySQL),

NoSQL (Redis)

EngineYard Cloud,

Orchestra PHP

Runs on EC2; enterprise version

runs on Terremark

Comments

LongJumpJava and JavaScript/ LongJump

FrameworkNoSQL (Proprietary)LongJump Cloud

Applications Platform

Runs on Rackspace; also sells

PaaS software separately

IBMNone; focused on tools for

deploying/managing appsRelational (DB2)

IBM SmartCloud

Application ServicesAnnounced October 2011;

not really a PaaS platform

sac-853t: inside windows azure: the cloud operating...

Documents