Deploying Exchange 2000 3

1
Deploying Exchange 2000Part 3 Storage and
Routing Paul BowdenProgram ManagerExchange
Server Product UnitMicrosoft Corporation
2
Session Theme

• Final part of the Exchange 2000 series
• Part 1 Directory
• Part 2 Co-existence and Upgrades
• Part 3 Storage and Routing
• Focus is on real deployments and ramifications,
  not feature sets

3
The Exchange 2000 Store
4
Terminology Buster Store
5
From 2000 Feet

• Scalability
• Multiple databases per server
• Front end/back end
• Performance
• Native content store
• Functionality
• Web Store
• Integrated Content Indexing
• Granular permissions

6
Multiple Databases

• Why?
• To improve backup/recovery times
• To classify different types of data
• Just to clarify
• Not designed for one mailbox per database
• We do not run multiple STORE.EXEs
• We can dynamically mount and dismount databases
• We can perform online restoration

7
Multiple MDB Scenarios

• Multiple companies on same server
• SLA conformance
• VIP mailboxes
• Large Public Folder applications
• Considerations
• Disk space
• Number of spindles
• Memory

8
Database Grouping

• Storage Groups
• Databases belong to a Storage Group
• Essentially, an ESE instance
• Each has a transaction log set
• Fail-over groups when clustering
• Multiples of (n-1) SGs per cluster

9
Storage Dilemmas

• New MDB or new Storage Group?
• Reasons for having a new group
• Circular logging requirements (e.g. NNTP feed)
• Different backup schedules
• Hard partition of company(s) data
• Clustering

10
Multi-MDB/SG Deployment
System Boot Partition (C)
HDD
HDD
Pagefile (D)
HDD
Database partition for SGs 1 and 2 (M)
Transaction Log Set 1 (E)
HDD
HDD
HDD
HDD
HDD
HDD
HDD
HDD
Transaction Log Set 2 (F)
HDD
HDD
Database partition for SGs 3 and 4 (N)
HDD
HDD
HDD
HDD
HDD
HDD
Transaction Log Set 3 (G)
HDD
HDD
Transaction Log Set 4 (H)
HDD
HDD
11
Clustering

• 2, 3 or 4 node clustering
• DataCenter Edition required for 4 node
• Clustering is Active Active
• Storage Groups are cluster resources
• Dynamic mount of storage group upon failure
• Run nodes at appropriate load
• 2 node 50 max load
• 3 node 66 max load
• 4 node 75 max load

12
Clustering Walkthrough
X
SG4
SG1
SG5
SG2
SG6
SG3
SG4
Disk Array
SG7
SG10
SG8
SG11
SG9
SG12
SG5
SG6
13
Client Access

• You dont need to upgrade clients to take
  advantage of multiple MDBs
• Client is transparent to Mailbox location
• Call to Mailbox namespace
• Mailboxes can be moved around
• Mixed-mode
• Anywhere within the same Admin Group
• Native-mode
• Anywhere within the forest

14
Backup And Recovery

• New NTBACKUP
• Enhanced Backup API
• Granularity is MDB level
• Single MDB backup backs up T. Logs
• Will need to backup entire storage group to flush
  logs
• Online MDB recovery
• Assumes Transaction Logs are intact
• Replays transactions with appropriate MDB ID

15
Front End / Back End

• Protocol and Database split
• Works with POP3, IMAP4, and HTTP
• Advantages
• Unified namespace
• Scalability / Load balancing

16
Front-End Deployment
ServerA
ServerC
ServerB
/exchange/pbowden
/exchange/larryl
/disc/XML
/disc/foo
/exchange/jkenerso
/disc/dogfood
/exchange/davidmad
Front End Servers
/disc/foo
Directory
/exchange/pbowden
17
Web Store

• In the past
• Where do I put my data?
• Public Folders
• Web Server
• File share
• Now
• Web Store

18
Content Indexing

• Enabled on a per-MDB basis
• Microsoft Search Service
• Transparent to clients
• Outlook will automatically use this data
• Will also search deep attachments and objects

19
Key Points

• Multiple MDBs provide great scalability
• Enhanced clustering
• Front-End servers provide a unified namespace for
  Internet clients
• Web Store and Content Indexing opens up huge
  possibilities
• No need for clients to be upgraded

20
Message Routing
21
Terminology Buster Routing
22
From 2000 Feet

• Performance
• SMTP native message transfer
• SMTP becomes a full peer of X.400
• Enhanced SMTP services
• Functionality
• New routing architecture
• Link State Algorithm
• Transport Event Sinks

23
Routing Groups

• Defines a set of meshed servers
• Similar to sites in Exchange Server 5.5
• All message transfer is SMTP-based

RGC
RGC
24
Planning Routing Groups

• Routing group design
• Resilient links required inside the RG
• Plan RGs on network bandwidth
• Identify traffic patterns
• Design ramifications
• No alignment with namespace
• No RPC involved
• Start with a 11 site to RG mapping
• RGs can be changed dynamically

25
Connector Options
RG2
RG1
Routing Group Connector Int. routing
SMTP Connector DNS routing
X.400 Connector Int. routing
26
Common SMTP Questions

• Why use SMTP?
• Common standard/better interoperability
• High-throughput
• Does SMTP mean bigger messages?
• Not for most messages
• Exchange will decide on the best transport
• Is SMTP slower than what I have today?
• No, transfer will be fast over all types of
  network links

27
More Questions

• Doesnt SMTP mean no security?
• Options for authentication and encryption
• What if I want compression?
• Easy to implement through Sinks

28
Link State Information

• Terminology
• LSA Link State Algorithm
• What is it?
• The new Routing mechanism
• Replaces the GWART
• Uses a mechanism similar to update sequence
  numbers

29
RG Master And Propagation

• Each RG has a Master
• Owns the table for the RG
• Immediate propagation of data
• Intra-RG propagation
• LSA uses TCP connection on port 3044
• Inter-RG propagation
• Uses X-LINK2STATE SMTP verb extension

30
Routing Walkthrough
RG1
RG2
10
X
10
20
RG3
X
RG5
20
10
RG4
Exchange 2000 Server
Infinite Cost? Wait in queue
RG Master
Routing Group Connector
31
Mixed-Vintage Routing
Ex5.5
Ex5.5
Ex5.5
Ex5.5
Ex5.5
Ex5.5
Ex5.5
Ex5.5
32
Event Sinks

• Programmatic access to SMTP/NNTP
• Easier to write than IMS extensions!
• Good documentation
• Can be written in any COM-compatible language
• Types
• Protocol
• Transport

33
Key Points

• SMTP is good
• High performance over all links
• Routing Groups are flexible
• Can change with the infrastructure
• Native connectors
• No more RPC worries
• No more ping-pong
• Event Sinks are very cool

34
Questions?
35
(No Transcript)
36
(No Transcript)