Grouping Technology In Windows NT Server, Enterprise Edition Jim Gray Microsoft Research GrayMicrosoft research.Micro .

Uploaded on:
Today\'s Agenda. Windows NT
Slide 1

Bunching Technology In Windows NT Server, Enterprise Edition Jim Gray Microsoft Research

Slide 2

Today\'s Agenda Windows NT ® grouping MSCS (Microsoft Cluster Server) Demo MSCS foundation Design objectives Terminology Architectural points of interest Setting up a MSCS bunch Hardware contemplations Cluster application issues Q&A

Slide 3

Extra Credit Included in your presentation materials yet not shrouded in this session Reference materials SCSI preliminary Speakers notes included Hardware Certification

Slide 4

MSCS In Action

Slide 5

High Availability Versus Fault Tolerance High Availability: veil blackouts through administration reclamation Fault-Tolerance: cover neighborhood issues RAID circles Uninterruptible Power Supplies Cluster Failover Disaster Tolerance: covers site disappointments Protects against flame, surge, attack,.. Excess framework and administration at remote site

Slide 6

Windows NT Clusters What is grouping to Microsoft? Gathering of autonomous frameworks that show up as a solitary framework Managed as a solitary framework Common namespace Services seem to be "group wide" Ability to endure segment disappointments Components can be added straightforwardly to clients Existing customer network is not affected by bunched applications

Slide 7

Microsoft Cluster Server 2-hub accessible 97Q3 Commoditize adaptation to non-critical failure (high accessibility) Commodity equipment (no extraordinary equipment) Easy to set up and oversee Lots of uses work out of the case. Multi-hub Scalability in NT5 time span

Slide 8

MSCA Initial Goals Manageability Manage hubs as a solitary framework Perform server upkeep without influencing clients Mask flaws, so repair is non-problematic Availability Restart fizzled applications and servers Un-accessibility ~ MTTR/MTBF , so fast repair Detect/caution executives of disappointments Reliability Accommodate equipment and programming disappointments Redundant framework without commanding a devoted "remain by" arrangement

Slide 9

MSCS Cluster Client PCs Server A Server B Heartbeat Cluster administration Disk bureau A Disk bureau B

Slide 10

Browser Server 1 Server 2 Failover Example Server 1 Server 2 Web website Web webpage Database Web website documents Database records

Slide 11

Basic MSCS Terms Resource - fundamental unit of failover Group - gathering of assets Node - Windows NT ® Server running bunch programming Cluster - at least one firmly coupled hubs, oversaw as a solitary substance

Slide 12

MSCS Namespace Cluster see Cluster name Node name Node name Virtual server name Virtual server name Virtual server name Virtual server name

Slide 13

MSCS Namespace Outside world view Cluster Node 1 Node 2 Virtual server 1 Virtual server 2 Virtual server 3 Internet Information Server SQL MTS "Bird of prey" Microsoft Exchange IP address: Network name: WHECCLUS IP address: Network name: WHECNode1 IP address: Network name: WHECNode2 IP address: Network name: WHEC-VS1 IP address: Network name: WHEC-VS2 IP address: Network name: WHEC-VS3

Slide 14

Windows NT Clusters Target applications Application & Database servers E-mail, groupware, efficiency applications server Transaction handling servers Internet Web servers File and print servers

Slide 15

MSCS Design Philosophy Shared nothing Simplified equipment setup Remoteable devices Windows NT sensibility improvements Never take a "bunch" down: shell amusement moving update Microsoft ® BackOffice™ item bolster Provide bunching answers for all levels of client necessities Eliminate cost and many-sided quality obstructions

Slide 16

MSCS Design Philosophy Availability is center for all discharges Single server picture for organization, customer collaboration Failover accommodated unmodified server applications, unmodified customers (group mindful server applications get wealthier components) Failover for record and print are default Scalability is stage 2 center

Slide 17

Non-Features Of MSCS Not bolt step/blame tolerant Not ready to "move" running applications MSCS restarts applications that are bombed over to other group individuals Not ready to recoup shared state amongst customer and server (i.e., document position) All customer/server exchanges ought to be nuclear Standard customer/server advancement controls still apply ACID dependably wins

Slide 18

Setting Up MSCS Applications

Slide 19

Attributes Of Cluster-Aware Applications A perseverance model that backings methodical state move Database case ACID exchanges Database log recuperation Client application bolster IP customers just How are retries upheld? No name benefit area conditions Custom asset DLL is something worth being thankful for

Slide 20

MSCS Services For Application Support Name benefit mapper GetComputerName makes plans to virtual server name Registry replication Key and fundamental keys and values are repeated to the next hub Atomic Logged to guarantee segments in time are dealt with

Slide 21

Application Deployment Planning System arrangement is essential Adequate equipment design You can\'t run Microsoft BackOffice on a 32-MB 75mhz Pentium Planning of favored gathering proprietors Good comprehension of single-server execution is basic See Windows NT Resource Kit execution arranging segment Understand working set size What is satisfactory execution to the specialty units?

Slide 22

Evolution Of Cluster-Aware Applications Active/inactive - general out-of-the-container applications Active/dynamic - applications that can run at the same time on different hubs Highly versatile - augmenting the dynamic/dynamic through I/O shipping, prepare bunches, and different methods

Slide 23

Application Evolution Application Node 1 Node 2 Microsoft SQL Server  Microsoft Transaction  Server (MTS) Internet Information  Server (IIS) Microsoft Exchange  Server

Slide 24

Evolution Of Cluster-Aware Applications Application Node 1 Node 2 Node 3 Node 4 Microsoft SQL Server     Microsoft Transaction  Server (MTS) Internet Information  Server (IIS) Microsoft Exchange  Server

Slide 25

Resources What are they? Assets are essential framework parts, for example, physical circles, forms, databases, IP addresses, and so forth., that give an administration to customers in a customer/server environment They are online in just a single place in the bunch at once They can bomb over starting with one framework in the bunch then onto the next framework in the group

Slide 26

Resources MSCS incorporates asset DLL bolster for: Physical and coherent plate IP address and system name Generic administration or application File share Print line Internet Information Server virtual roots Distributed Transaction Coordinator (DTC) Microsoft Message Queue (MSMQ) Supports asset conditions Controlled through very much characterized interface Group: offers a "virtual server"

Slide 27

Cluster Service To Resource Windows NT bunch benefit Resource screen Initiate changes Resource occasions Physical plate asset DLL IP address asset DLL Generic application asset DLL Database asset DLL Disk Network App Database

Slide 28

Cluster Abstractions Resource Cluster Resource Group Resource: program or gadget oversaw by a group e.g., document benefit, print benefit, database server can rely on upon different assets (startup requesting) can be on the web, disconnected, delayed, fizzled Resource Group: an accumulation of related assets has assets; has a place with a group unit of co-area; required in naming assets Cluster: a gathering of hubs, assets, and gatherings collaboration for confirmation, organization, naming

Slide 29

Resources have... Sort: what it files, (DB, print, Web… ) An operational state (on the web/disconnected/fizzled) Current and conceivable hubs Containing Resource Group Dependencies on different assets Restart parameters (if there should be an occurrence of asset disappointment) Resource Cluster Group

Slide 30

Resource Fails over (moves) starting with one machine then onto the next Logical circle IP address Server application Database May rely on upon another asset Well-characterized properties controlling its conduct

Slide 31

Resource Dependencies An asset may rely on upon different assets An asset is brought online after any assets it relies on upon An asset is taken disconnected before any assets it relies on upon All needy assets must bomb over together

Slide 32

Dependency Example Database asset DLL Generic application asset DLL IP address asset DLL Drive E: asset DLL Drive F: asset DLL

Slide 33

Group Example Payroll gather Database asset DLL Generic application asset DLL IP address asset DLL Drive E: asset DLL Drive F: asset DLL

Slide 34

Resource screens MSCS Architecture Cluster API Cluster overseer Cluster API DLL Cluster API stub Cluster.Exe Cluster API DLL Global Update Manager Log Manager Database Manager Membership Manager Event Processor Checkpoint Manager Object Manager Node Manager Failover Manager Resource Manager Application asset DLL Resource API Physical asset DLL Logical asset DLL Application asset DLL Reliable Cluster Transport + Heartbeat Network

Slide 35

MSCS Architecture Cluster administration is involved the accompanying articles Failover Manager (FM) Resource Manager (RM) Node Manager (NM) Membership Manager (MM) Event Processor (EP) Database Manager (DM) Object Manager (OM) Global Update Manager (LM) Checkpoint Manager (CM) More about these in the following session

Slide 36

Setting Up A MSCS Cluster

Slide 37

MSCS Key Components Two servers Multi versus uniprocessor Heterogeneous servers Shared SCSI transport SCSI HBAs, SCSI RAID HBAs, HW RAID boxes Interconnect Many sorts can be bolstered Remember, two NICs per hub PCI for bunch interconnect Complete MSCS HCL design

Slide 38

MSCS Setup Most regular issues Duplicate SCSI IDs on connectors Incorrect SCSI c

View more...