1

Windows NT Scalability

Jim Gray

Microsoft ResearchGray@Microsoft.com

http/www.research.Microsoft.com/~Gray/talks/

2

OutlineOutline

• Scalability: What & Why?• Scale UP: NT SMP scalability• Scale OUT: NT Cluster scalability• Key Message:

– NT can do the most demanding apps today.

– Tomorrow will be even better.

Scale OutScale Out

Scale Up

Scale DownDown

3

Scale OutScale Out

Server ClusterServer ClusterWhat is Scalability?

• Grow without limits– Capacity– Throughput– Availability

• Do not add complexity– design– administer– Operate– UseS

cale

Do

wn

Do

wn

Win TermWin TermNetPCNetPC

HandheldHandheld

PortablePortable

TVTV

Sca

le U

pSuperSuperServerServer

ServerServer

PC PC WorkstationWorkstation

4

Scalability is Important• Automation benefits rising

– ROI of 1 month....

• Slice price going to zero– Cyberbrick costs 5k$

• Design, Implement & Manage cost going down

– Internet tools, Database tools– DCOM & Transactions, clusters.

• Billions of clients imply millions of HUGE servers.

• Thin clients imply huge servers.

ServerServer

5

1987: 256 tps $ 14 million computerA dozen peopleTwo rooms of machines

1997: 1,250 tps $ 50 k$ computerOne person1 micro-dollar per transaction (1,000x cheaper)

What Happens in 10 Years?

Ready for the next 10 years?

6

Q: Why Does Microsoft Care? A: Billions of clients need millions of servers

Microsoft wants to do a GREAT! job on Scaleable Windows NT and Scaleable BackOffice.

Key technique: INTEGRATION.

0300600900

1,2001,5001,8002,1002,4002,700

1994 1995 1996 1997 1998 1999 2000 2001

UnixUnix

WindowsNT WindowsNT ServerServer

NetWareNetWare

Servers Shipped per year

(97-01 are MS estimates)

7

OutlineOutline

• Scalability: What & Why?• Scale UP: NT SMP scalability• Scale OUT: NT Cluster scalability

Scale OutScale Out

Scale Up

Scale DownDown

8

How Scaleable is NT??The Single Node Story

• 64 bit file system in NT 1, 2, 3, 4, 5• 8 node SMP in NT 4.E, 32 node OEM• 64 bit addressing in NT 5• 1 Terabyte SQL Databases (PetaByte capable)

• 10,000 users (TPC-C benchmark)• 100 Million web hits per day (IIS)• 50 GB Exchange mail store

next release designed for 16 TB

• 50,000 POP3 users on Exchange (1.8 M messages/day)

• And, more coming…..

9

Windows NT ServerEnterprise Edition

• Scalability – 8x SMP support (32x in OEM kit)– Larger process memory (3GB Intel)– Unlimited Virtual Roots in IIS (web)

• Transaction– DCOM transactions (Viper TP mon) – Message Queuing (Falcon)

• Availability– Clustering (WolfPack)– Web, File, Print, … servers fail over.

10

Manageability Windows NT 5.0 and Windows 98

• Active Directory tracks all objects in net• Integration with IE 4.

–Web-centric user interface• Management Console

–Component architecture• Zero Admin Kit and Systems

Management Server• Plug&Play, Instant On, Remote Boot,..• Hydra and Intelli-Mirroring

11

Windows NT ServerWindows NT Serverwith “Hydra” Serverwith “Hydra” Server

Dedicated Dedicated Windows Windows terminalterminal

Existing, Existing, Desktop PC Desktop PC

MS-DOS, MS-DOS, UNIX, UNIX, Mac Mac clientsclients

Net PCNet PC

Thin Client Supportlower per-client costs

12

The benefits of the PC plus The benefits of the PC plus centralized computing advantagescentralized computing advantages

Windows NT 5.0IntelliMirror™

• Files and settings mirrored on client and server

• Great for disconnected users• Facilitates roaming• Easy to replace PCs• Optimizes network performance

13

Scaleup To Big Databases?• NT 4 and SQL Server 6.5

– DBs up to 1 Billion records, – 100 GB– Covers most (80%) data warehouses

• SQL Server 7.0 – Designed for Terabytes

• Hundreds of disks per server.• SMP parallel search

– Data Mining and Multi-Media– Multi-Media is probably main

• TerraServer is good example

TTBB

ExcelExcelspreadsheetspreadsheet

Manhattan phone book Manhattan phone book (15MB)(15MB)

Human GenomeHuman Genome (3GB) (3GB)

Dayton-HudsonDayton-HudsonSales recordsSales records(300GB)(300GB)

SatelliteSatellitephotos of photos of

Earth (1 TB)Earth (1 TB)

14

Database ScaleUP: TerraServer™• Build a 1 TB SQL Server database

– Demo NT and SQL Server scalability– Stress test SQL Server 7.0

• Data must be– 1 TB– Unencumbered– Interesting to everyone everywhere– And not offensive to anyone anywhere

• Loaded – 1.1 M place names from Encarta World Atlas– 1 M Sq Km from USGS (1 meter resolution)– 2 M Sq Km from Russian Space agency (2 m)

• Will be on web (world’s largest atlas)• Sell images with commerce server.• USGS CRDA: 3 TB more coming.

15

TerraServer System

• DEC Alpha 4100 (4x smp) + • 324 StorageWorks Drives (1.4 TB)• RAID 5 Protected• SQL Server “Sphinx”• USGS 1-meter data

(30% of US)• Russian Space data

Two meterresolutionimages(2 M km2 2% of earth)

SPIN-2SPIN-2

16

Demo by Tom Barclay

Http://t2b2

17

How Can We Measure Scalability?• SpecMarks™ ?

–Just measures processor• Transactions per Minute

– Transaction Processing Performance Council– Measures whole system– Price and Performance– 63 data points

• 3 year history• all vendors

– audited results– www.tpc.org

tpmC vs Time

05,000

10,00015,00020,00025,00030,00035,000

Jan-95 Jan-96 Jan-97

tpm

C

18

NT vs UNIX SMPs• NT typically runs on 1 to 4 cpus

– Scales near-linear on them• UNIX boxes: 32-64 way SMPs

– They do 3x more tpmC– They cost 10x more.

• 10 way NT machines are available– They cost more– They are faster

• My view (shared by many)– Need clusters for availability– Cluster commodity servers to make huge systems– a la Tandem, Teradata, VMScluster, IBM Sysplex, IBM SP2

tpmC vs Time

05,000

10,00015,00020,00025,00030,00035,000

Jan-95 Jan-96 Jan-97

tpm

C

tpmC vs Time

05,000

10,00015,00020,00025,00030,00035,000

Jan-95 Jan-96 Jan-97

tpm

C

hh

UnixNT

tpmC vs Time

05,000

10,00015,000

20,00025,000

30,00035,000

Jan-95 Jan-96 Jan-97

tpm

C

hUnix

NT

19

Transaction Throughput TPC-C• On comparable hardware, NT scales best• SQL Server & NT Improving 250% per year• Best Price Performance (2x cheaper)tpmC on Intel CPUs

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

0 1 2 3 4 5 6 7 8 9 10

tpm

C

NT

UNIX

h

h

hhhh

tpmC vs Intel CPUs

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

0 1 2 3 4 5 6 7 8 9 10

tpm

C

NT all

NT Best

Unix best

hh

20

NT Scales Better Than Solaris• Microsoft SQL

NTIntel scales to 6x

• Beats Sybase Solaris UltraSPARCup to 11-way

0

5,000

10,000

15,000

20,000

0 10 20cpus

tpm

C

Sybase/Solaris

/UltraSPARC

MS

SQL/N

T/Inte

l

21

Only NT Has Economy of Scale

• NT is 2x less expensive

• Only NT has economy of scale

• Unix has dis-economy of scale

Transactions/k$ by vendor

0.0

5.0

10.0

15.0

20.0

25.0

0 10,000 20,000 30,000tpmC

tpm

C/k

$

DB2/UnixSybase/Unix

Informix/Unix

Microsoft/NT

Oracle/Unix

22

What Does This Mean?

NT

10,000 usersLess than 10,000 usersuse NT

?

20,000 users

?

30,000 users10k-30K users

consider UNIXrequire high availability

=> clusters (scale out)

NT*

NT*

40,000 users

NT*

50,000 users30k-1M users require cluster (scale out)require high availability (clusters)

NT*NT*NT*NT*NT*

NT*NT*NT*NT*

NT*NT*NT*

23

What Does This Mean?

Less than 10,000 usersuse NT

10k-30K usersconsider UNIXrequire high availability

=> clusters (scale out)

30k-1M band require cluster (scale out)require high availability (clusters)

Need h

igh-a

vaila

bility

clu

ster

Need h

igh-a

vaila

bility

clu

ster

even

at 1

0,00

0 use

rs

even

at 1

0,00

0 use

rs

10,000 users

20,000 users

30,000 users

40,000 users

50,000 users

24

OutlineOutline

• Scalability: What & Why?• Scale UP: NT SMP scalability• Scale OUT: NT Cluster scalability• Key Message:

– NT can do the most demanding apps today.

– Tomorrow will be even better.

Scale OutScale Out

Scale Up

Scale DownDown

25

Clusters Have Advantages• High Availability: Fault tolerance:

– Spare modules mask failures

• Modular growth– Grow by adding small modules

• Parallel data search– Use multiple processors and disks

• Clients and servers made from the same stuff– Inexpensive: built with commodity CyberBricks

26

How scaleable is NT??The Cluster Story

• 16-node Tandem Cluster– 64 cpus– 2 TB of disk– Decision support

• 45-node Compaq Cluster– 140 cpus– 14 GB DRAM– 4 TB RAID disk– OLTP (Debit Credit)

• 1 B tpd (14 k tps)

27

microsoft.com

• Production

– Windows NT V4 and IIS V3• 20 HTTP, • 3 download, • 3 FTP• 5 SQL 6.5• Index Server + 3 search

• Stagers– Site Server for content– DCOM Publishing wizard

• Network– 4 100MB FDDI rings– 6 DS3– 4 TB/day download capacity

• Replicas in UK and Japan

• 90m hits/day– 16.5m page views– #4 site on Internet

• 900k visitors per day• Not cheap

– Data Centers– Bandwidth– 27 people on content – 22 people on systems

28

Tandem 2 Ton

• 2 TB SQL database • 1.2 TB user data• 16 node cluster• 64 cpus, 480 disks• Decision support

parallel data-mining

• Will be Wolf Pack aware• Demoed at DB Expo in• ServerNet™ interconnect

29

Billion Transactions per Day Demo

• Built a 45-node Windows NT Cluster (with help from Intel & Compaq) > 900 disks, 3 TB

• All off-the-shelf parts• Using SQL Server &

DTC distributed transactionsDCOM & ODBC clientson 20 front-end nodes

• DebitCredit Transaction• Each server node has 1/20 th of the DB • Each server node does 1/20 th of the work• 15% of the transactions are “distributed”

30

Cluster Architecture

VIPDTC1 VIPDTC2 VIPDTC3 VIPDTC4 VIPDTC5

ControlDriver

DatabaseSwitchDTC

VIPDC42 VIPDC43 VIPDC44 VIPDC45 VIPDC46 VIPDC47 VIPDC48 VIPDC49 VIPDC50 VIPDC51

VIPDC2 VIPDC3 VIPDC4 VIPDC5 VIPDC6 VIPDC7 VIPDC8 VIPDC9 VIPDC10 VIPDC11

VIPDC12 VIPDC13 VIPDC14 VIPDC15 VIPDC16 VIPDC17 VIPDC18 VIPDC19 VIPDC20 VIPDC21

31

1.2 B tpd• 1 B tpd ran for 24 hrs.• Out-of-the-box software• Off-the-shelf hardware• AMAZING!

• Sized for 30 days• Linear growth• 5 micro-dollars per transaction

32

Millions of Transactions Per Day

0.1

1.

10.

100.

1,000.

1 Btpd Visa ATT BofA NYSE

Mtp

d

Millions of Transactions Per Day

0.100.200.300.400.500.600.700.800.900.

1,000.

1 Btpd Visa ATT BofA NYSE

Mtp

d

How Much Is 1 Billion Tpd?• 1 billion tpd = 11,574 tps

~ 700,000 tpm (transactions/minute)• ATT

– 185 million calls per peak day (worldwide)

• Visa ~20 million tpd– 400 million customers– 250K ATMs worldwide– 7 billion transactions

(card+cheque) in 1994 • New York Stock Exchange

– 600,000 tpd• Bank of America

– 20 million tpd checks cleared (more than any other bank)– 1.4 million tpd ATM transactions

• Worldwide Airlines Reservations: 250 Mtpd

33

Cluster Server: High Availability• Multiple servers form one system• Industry standard APIs and hardware• Server application and tools support

– IIS web server– File and Print servers– IP and NetName failover– Transaction and Queue Server failover– SQL Server, Enterprise edition

• Two-Node clusters now (2 to 20 cpus)• 16 node soon (2 to 192 cpus).

34

Web Web sitesite

DatabaseDatabase

Web site filesWeb site files

Database filesDatabase files

BrowserBrowser

HP-WolfPack ClusterIIS & SQL High Availability

Web Web sitesite

DatabaseDatabase

AliceAlice BettyBetty

35

SQL Server and IISSymmetric Virtual Servers

Web Web sitesite

DatabaseDatabase

Web site filesWeb site files

Database filesDatabase files

BrowserBrowser

Web Web sitesite

DatabaseDatabase

AliceAlice BettyBetty

Web site filesWeb site files

Database filesDatabase files

• Both Servers deliver service.• Almost no overhead during normal operation.• Each protects the other.• High availability:

– Planned outages for maintenance

– Unplanned outages (failures)

36

3 Tier WebWindows Windows TerminalTerminal BrowserBrowser NetPCNetPC WindowsWindows OtherOther

Data Stores: SQL, Messaging, ISAM, Host, Unstructured StorageData Stores: SQL, Messaging, ISAM, Host, Unstructured Storage

A consistent programming model for integrating all data, services and componentsA consistent programming model for integrating all data, services and components

““K2”K2”Services: Transactions, Security, Message QueuingServices: Transactions, Security, Message Queuing

COM and other standard protocolsCOM and other standard protocols

COM and other standard protocolsCOM and other standard protocols

ASP

ASP

ASP

ObjectObject

Use ASPs for presentation

Use COM for Business

Rules

37

Data Access Within NTWindows Windows TerminalTerminal BrowserBrowser NetPCNetPC WindowsWindows OtherOther

TransactionTransaction

ODBCODBCResourceResource

Dispenser Dispenser

SQLSQLServerServer

SQLSQLServerServer

Object

Object

38

Cedar: Accessing Legacy DataWindows Windows TerminalTerminal BrowserBrowser NetPCNetPC WindowsWindows OtherOther

TransactionTransaction

ODBCODBC

SQLSQLServerServer

SQLSQLServerServer

CedarCedar

IMSIMS CICSCICS

SNA ServerSNA Server

LU 6.2LU 6.2

Object

Object

39

MSMQ: Queued Processing Windows Windows TerminalTerminal BrowserBrowser NetPCNetPC WindowsWindows OtherOther

TransactionTransaction

ODBCODBC

SQLSQLServerServer

SQLSQLServerServer

CedarCedar

IMSIMS CICSCICS

SNA ServerSNA Server

MSMQMSMQ

MSMQMSMQ

LU 6.2LU 6.2

Object

Object

40

Accessing UNIX DBMS’sWindows Windows TerminalTerminal BrowserBrowser NetPCNetPC WindowsWindows OtherOther

TransactionTransaction

ODBCODBC

SQLSQLServerServer

SQLSQLServerServer

CedarCedar

IMSIMS CICSCICS

SNA ServerSNA Server

MSMQMSMQ

MSMQMSMQ DB2, OracleDB2, OracleInformixInformix

XAXALU 6.2LU 6.2

Object

Object

Third parties coming

41

DCOM Completes the PictureWindows Windows TerminalTerminal BrowserBrowser NetPCNetPC WindowsWindows OtherOther

Data Stores: SQL, Messaging, ISAM, Host, Unstructured StorageData Stores: SQL, Messaging, ISAM, Host, Unstructured Storage

A consistent programming model for integrating all data, services and componentsA consistent programming model for integrating all data, services and components

““K2”K2”Services: Transactions, Security, Message QueuingServices: Transactions, Security, Message Queuing

COM and other standard protocolsCOM and other standard protocols

COM and other standard protocolsCOM and other standard protocols

ASP

ASP

ASP

ObjectObject

Object

Object

Object

42

•20% Performance Advantage, Cluster Aware20% Performance Advantage, Cluster Aware•OLAP Support, Natural Language InterfaceOLAP Support, Natural Language Interface•Available: 97Q3 (US), Loc: F, G, Japanese Available: 97Q3 (US), Loc: F, G, Japanese

•Merchant ServerMerchant Server•Usage analyst Usage analyst •Available: 97Q3 (US)Available: 97Q3 (US)

•All Enterprise EditionsAll Enterprise Editions•Multi-ServerMulti-Server•Available: 97Q4Available: 97Q4

•Designed for Terabyte message stores Designed for Terabyte message stores •Cluster AwareCluster Aware•Available 97Q4 (US)Available 97Q4 (US)

2.0/E2.0/E

2.5/E2.5/E

xx /Exx /E

6.5/E6.5/E

BackOffice Enterprise Road Map

43

SummarySummary

• Scalability: What & Why?• Scale UP: NT SMP scalability• Scale OUT: NT Cluster scalability• Key Message:

– NT can do the most demanding apps today.

– Tomorrow will be even better.

Scale OutScale Out

Scale Up

Scale DownDown

44

Questions?

• Slides at:

– http://www.Research.Microsoft.com/~Gray

• I will be on show floor: 11:00: Microsoft booth

Mail to Gray@Microsoft.com

45

Cluster-Wide Load Balance

Package A Package BPackage B Package CPackage C Package A Package A Package A

Static Load BalancingStatic Load Balancing

IISPackage APackage A IIS Package APackage A

DNS Round RobinDNS Round Robin Dynamic Load BalancingDynamic Load Balancing

Package A Package A Package A