SharePoint Global Deployment Joel Oleson Sr. Technical Product Manager http://sharepointjoel.com Session Code:

7+ Year SharePoint Veteran 5 Years in MS IT 2 Years in Product Team SharePoint Consultant and Trainer 5 yrs TechEd US & Europe Globe Trotter Ate crickets last week Who is SharePoint Joel?

7+ Year SharePoint Veteran

5 Years in MS IT

2 Years in Product Team

SharePoint Consultant and Trainer

5 yrs TechEd US & Europe

Globe Trotter

Ate crickets last week

Objectives And Takeaways Design Geo-distributed SharePoint Deployments O ptimize the user experience Deal with network latency and improve the general performance of the solution.

Design Geo-distributed SharePoint Deployments

O ptimize the user experience

Deal with network latency and improve the general performance of the solution.

Agenda Geo Dilemma Deployment Options Information Architecture Search Architecture for the WAN Optimizing for the WAN WAN Performance testing

Geo Dilemma

Deployment Options

Information Architecture

Search Architecture for the WAN

Optimizing for the WAN

WAN Performance testing

Geo Dilemma To distribute or not distribute is the question Cost vs. Usability Bandwidth can be bought but latency is bound to laws of physics Different stakeholders -> different priorities/preferences IT tends to favor Centralized (low cost) Business just wants it to work (faster the better)

Cost vs. Usability

Bandwidth can be bought but latency is bound to laws of physics

Different stakeholders -> different priorities/preferences

IT tends to favor Centralized (low cost)

Business just wants it to work (faster the better)

Agenda Geo Dilemma Deployment Options Information Architecture Search/Indexing Architecture for the WAN Optimization Techniques WAN Performance Testing

Geo Dilemma

Deployment Options

Information Architecture

Search/Indexing Architecture for the WAN

Optimization Techniques

WAN Performance Testing

Deployment Options Central solution Regional solution Distributed solution

Central solution

Regional solution

Distributed solution

Central Solution

Central Deployment with a twist Partner Solution: WAN Acceleration REDMOND WAN Accelerator Datacenter All Services in one Central Farm Central Search Central Directory WAN Accelerator remote office BEIJING 10s-100s of Local WAN Accelerators ~5x - 1 st Request ~43x - 2 nd Request

Sampling of WAN Optimization Partners WAN Accelerators Cisco Citrix Certeon Packeteer Riverbed F5 Cache Devices Cisco F5 Inktomi ISA and IAG Server Offline Client Solutions Groove Colligo Data Replication Syntergy WinApp Technologies Casahl Infonic Multi-farm management Quest Software EchoTechnologies IDevFactory AvePoint CorasWorks Barracuda Tools CommVault Symantec Byte Level Replica Neverfail Double-take SAN Replication HP EMC Centera Hitachi Data Systems

WAN Accelerators

Cisco

Citrix

Certeon

Packeteer

Riverbed

F5

Cache Devices

Cisco

F5

Inktomi

ISA and IAG Server

Offline Client Solutions

Groove

Colligo

Data Replication

Syntergy

WinApp Technologies

Casahl

Infonic

Multi-farm management

Quest Software

EchoTechnologies

IDevFactory

AvePoint

CorasWorks

Barracuda Tools

CommVault

Symantec

Byte Level Replica

Neverfail

Double-take

SAN Replication

HP

EMC Centera

Hitachi Data Systems

Central Solution Easiest to maintain and operate. Cheapest Solution Easiest to support Is it the Utopia? - Don’t forget D/R!!!

Easiest to maintain and operate.

Cheapest Solution

Easiest to support

Is it the Utopia? - Don’t forget D/R!!!

Key Network Considerations Latency #1 Bandwidth #2 Utilization (is it already high % consumed?) Expected use of MOSS (page sizes, file sizes, and usage patterns)

Latency #1

Bandwidth #2

Utilization (is it already high % consumed?)

Expected use of MOSS (page sizes, file sizes, and usage patterns)

Other Network Considerations Politics (AD trusts/Affiliates) Deloitte Business Requirements (Speed comfort) Energizer Network Map (Microsoft) Network Roadmap & Datacenter Consolidation Goals (HP)

Politics (AD trusts/Affiliates) Deloitte

Business Requirements (Speed comfort) Energizer

Network Map (Microsoft)

Network Roadmap & Datacenter Consolidation Goals (HP)

Regional Deployment Optimized Network Bandwidth/Latency REDMOND DUBLIN SINGAPORE Regional Scope Services Local Office Server Farms (Intranet only) Local SSP Farm Centrally Managed from Redmond Enterprise Scope Services Local Office Server Farms (Intranet and Extranet) Local SSP Farm Centrally Managed from Redmond Regional Scope Services Local Office Server Farms (Intranet and Extranet) Local SSP Farm Centrally Managed from Redmond

Regional Scope Services

Local Office Server Farms (Intranet only)

Local SSP Farm

Centrally Managed from Redmond

Enterprise Scope Services

Local Office Server Farms (Intranet and Extranet)

Local SSP Farm

Centrally Managed from Redmond

Regional Scope Services

Local Office Server Farms (Intranet and Extranet)

Local SSP Farm

Centrally Managed from Redmond

Regional Deployments Central site hosts for regional sites Regional sites host Server farm running MOSS including: Published intranet content. Enterprise-wide search (optional). My Site redirection (Trusted My Site Host Location). This ensures that a user coming from a regional farm is redirected to the My Site on the farm where their profile resides. Profile synchronization. Aggregation of content developed at regional sites based on workflow and policies. Server farm running MOSS, including: Team or division collaboration sites. My Sites with My Site redirection. This ensures that a user coming from the central farm or another regional farm is redirected to the My Site where their profile resides. Search for local content.

Server farm running MOSS including:

Published intranet content.

Enterprise-wide search (optional).

My Site redirection (Trusted My Site Host Location). This ensures that a user coming from a regional farm is redirected to the My Site on the farm where their profile resides.

Profile synchronization.

Aggregation of content developed at regional sites based on workflow and policies.

Server farm running MOSS, including:

Team or division collaboration sites.

My Sites with My Site redirection. This ensures that a user coming from the central farm or another regional farm is redirected to the My Site where their profile resides.

Search for local content.

Why Go Regional? If you are unable to optimize the central solution to improve WAN performance to acceptable levels. Provides fast access to services that regional users are likely to use most —collaboration sites and My Sites. Provides separate deployments for geo-political or compliance related

If you are unable to optimize the central solution to improve WAN performance to acceptable levels.

Provides fast access to services that regional users are likely to use most —collaboration sites and My Sites.

Provides separate deployments for geo-political or compliance related

Why Not Go Regional? Regional provides more overhead and backup complexity – more environments to support Difficult to get Indexing to work as you’d want it to over the WAN (consumes Bandwidth) Shared Services such as BDC and Profiles can’t be shared

Regional provides more overhead and backup complexity – more environments to support

Difficult to get Indexing to work as you’d want it to over the WAN (consumes Bandwidth)

Shared Services such as BDC and Profiles can’t be shared

Distributed- Multiple Deployments with Consolidation Search where possible Denver HQ Central Portal MOSS farm for Enterprise Search Branch Office MOSS or WSS Deployments BANGALORE Disconnected or Bandwidth Constrained or Autonomy

Distributed Central site hosts for regional sites Local/Distributed sites host Server farm running MOSS, including: Published intranet content Aggregation of some local content My Site redirection and profile synchronization if MOSS is used at the regional sites. Either: Server farm running MOSS, including: Team collaboration Regional or Aggregated My Sites Search for local content Read-only copy of published intranet content (optional) Server farm running WSS including: Team or division collaboration sites. WSS Search

Server farm running MOSS, including:

Published intranet content

Aggregation of some local content

My Site redirection and profile synchronization if MOSS is used at the regional sites.

Either:

Server farm running MOSS, including:

Team collaboration

Regional or Aggregated My Sites

Search for local content

Read-only copy of published intranet content (optional)

Server farm running WSS including:

Team or division collaboration sites.

WSS Search

Why go Distributed? Recommended for organizations that have many offices distributed geographically, such as branch offices. Fastest possible access to most remote users Bandwidth challenged Autonomous organizations can find MOSS Central brings distributed deployments together

Recommended for organizations that have many offices distributed geographically, such as branch offices.

Fastest possible access to most remote users

Bandwidth challenged

Autonomous organizations can find MOSS Central brings distributed deployments together

Why Not? It’s the easiest to decide (because the business wants fast), but hardest to get working right Most amount of support Most complex for operations Can be confusing for users with cross group/cross location collaboration

It’s the easiest to decide (because the business wants fast), but hardest to get working right

Most amount of support

Most complex for operations

Can be confusing for users with cross group/cross location collaboration

Determining Latency and Page and File Elements Simple network tools… Using the power of Ping Wfetch and Fiddler

Key Considerations… Reinforced. Know your usage Intranet publishing portal Collaboration BI, reporting Know your network Bandwidth Latency Utilization Include tests and Surveys* from remote offices

Know your usage

Intranet publishing portal

Collaboration

BI, reporting

Know your network

Bandwidth

Latency

Utilization

Include tests and Surveys* from remote offices

Agenda Geo Dilemma Deployment Options Information Architecture Search Architecture for the WAN Optimizing for the WAN WAN Performance Testing

Geo Dilemma

Deployment Options

Information Architecture

Search Architecture for the WAN

Optimizing for the WAN

WAN Performance Testing

Plan for Information Management Usage profiles Company/enterprise wide use Team use Personal use Life Cycle Cross X Collaboration Publishing-Content is authored by few and shared broadly. Archive – infrequent to no access

Usage profiles

Company/enterprise wide use

Team use

Personal use

Life Cycle

Cross X Collaboration

Publishing-Content is authored by few and shared broadly.

Archive – infrequent to no access

Plan for Global Collaboration Centralized vs. Distributed Global or local teams Where to put the sites End user experience Bandwidth/Latency Operational costs Search Offline Groove, Outlook, etc. Network accelerators Security, ACLs

Centralized vs. Distributed

Global or local teams

Where to put the sites

End user experience

Bandwidth/Latency

Operational costs

Search

Offline

Groove, Outlook, etc.

Network accelerators

Security, ACLs

Agenda Geo Dilemma Deployment Options Information Architecture Search Architecture for the WAN Optimizing for the WAN WAN Performance Testing

Geo Dilemma

Deployment Options

Information Architecture

Search Architecture for the WAN

Optimizing for the WAN

WAN Performance Testing

Global Scale Search Centralized vs. Distributed farms Crawling across the wire vs. Querying across the wire Crawling external content Protocol handlers iFilters Partitioning data Based on Geography vs. Content Federating results Merging results from different indexes loses relevancy

Centralized vs. Distributed farms

Crawling across the wire

vs.

Querying across the wire

Crawling external content

Protocol handlers

iFilters

Partitioning data

Based on Geography vs. Content

Federating results

Merging results from different indexes loses relevancy

Search architectures for WAN environments Centralized Search Regionalized Search Distributed Search Federated Search

Centralized Search

Regionalized Search

Distributed Search

Federated Search

Search architectures for WAN environments Centralized search The central SSP crawls all the content in the local farm Regional queries are sent to the central farm Pros: Centralized management World-wide relevancy Cons: Crawling/Querying across the wire Index freshness if regional content changes frequently Need manual redirection

The central SSP crawls all the content in the local farm

Regional queries are sent to the central farm

Pros:

Centralized management

World-wide relevancy

Cons:

Crawling/Querying across the wire

Index freshness if regional content changes frequently

Need manual redirection

Regionalized Search Each region hosts its own SSP Each SSP only crawls local content Search URLs for each Region Pros: No crawling across the wire Cons: Search is regionalized Users must know which farm to search

Each region hosts its own SSP

Each SSP only crawls local content

Search URLs for each Region

Pros:

No crawling across the wire

Cons:

Search is regionalized

Users must know which farm to search

Distributed Search Localized Central Site crawls Local Farms (where possible) Pros: Central has global relevancy Regional users can view regional results Cons: local users must use Central Search for global search Dual crawling or more…

Localized

Central Site crawls Local Farms (where possible)

Pros:

Central has global relevancy

Regional users can view regional results

Cons:

local users must use Central Search for global search

Dual crawling or more…

Federated Search Each region hosts its own SSP Each region configured to crawl itself and Federated Search Queries to other regions SSP is configured to crawl other farms based on the business needs MOSS2007 SP1 + Infra Update Pros: Query performance optimized locally Content is crawled once Asynchronous queries over WAN Cons: Relevancy is weighted Multiple SSPs to manage

Each region hosts its own SSP

Each region configured to crawl itself and Federated Search Queries to other regions

SSP is configured to crawl other farms based on the business needs

MOSS2007 SP1 + Infra Update

Pros:

Query performance optimized locally

Content is crawled once

Asynchronous queries over WAN

Cons:

Relevancy is weighted

Multiple SSPs to manage

Agenda Geo Dilemma Deployment Options Information Architecture Search Architecture for the WAN Optimizing for the WAN WAN Performance Testing

Geo Dilemma

Deployment Options

Information Architecture

Search Architecture for the WAN

Optimizing for the WAN

WAN Performance Testing

End-to-End Perceived Network Performance Client Performance: CPU/Processor Delay Virtual/Physical Memory Limits NIC Config, Network Drivers and TCP/IP Config Client Application Versions Proxy Config Client-side LAN Performance: Switch Config Contention for Uplink WAN: Link Speed(Bandwidth) Delay (Latency) Packet Loss, Retransmissions, Timeouts Network Throughput Round Trips Concurrency Server-Side LAN Performance: Switch Config Contention for Uplink Server Performance: CPU/Processor Delay Virtual and Physical Memory Limits NIC/Network Drivers and TCP/IP config Network Speed (10/100/1000MB) Server Software Layer

Client Performance:

CPU/Processor Delay

Virtual/Physical Memory Limits

NIC Config, Network Drivers and TCP/IP Config

Client Application Versions

Proxy Config

Client-side LAN Performance:

Switch Config

Contention for Uplink

WAN:

Link Speed(Bandwidth)

Delay (Latency)

Packet Loss, Retransmissions, Timeouts

Network Throughput

Round Trips

Concurrency

End to End Examples

Crawling has IMPACT Commonly 50% of Traffic Use Dedicated WFE for crawling; remove from NLB rotation Combine Index Server with Web server role (hosts)

Commonly 50% of Traffic

Use Dedicated WFE for crawling; remove from NLB rotation

Combine Index Server with Web server role (hosts)

SharePoint to SharePoint Crawl

File Server Crawl Rates

HTTP Crawl Rates

Intranet Recommended Bandwidth Centralize Regional Deployments Distributed Bandwidth ranges of 3Mbps (Dual T1) and above with latencies no greater than ~150 milliseconds with large offices (well connected) Bandwidth ranges of 1.5Mbps (T1) and above with latencies no greater than 100 msecs (*Well connected with hubs) Disconnected and/or limited bandwidth with large groups of users or no AD trust organizations (*non trusting disconnected, autonomous or high speed requirements)

What Do SharePoint Server and Donald Trump Have in Common? Courtesy Si.com

Cache! TechNet: (Cache Settings) Additional performance and capacity factors

Cache

Cache Config Levels Web App – Disk based caching in web.config Site collection – configure output cache and object cache settings Site – output cache settings Page layout – output cache Web Part – settings in dwp code Query – i.e. RSS Feed cache is 5 min by default, cross list query

Web App – Disk based caching in web.config

Site collection – configure output cache and object cache settings

Site – output cache settings

Page layout – output cache

Web Part – settings in dwp code

Query – i.e. RSS Feed cache is 5 min by default, cross list query

Cache Recommendations Cache is but…. Setting memory based caching can waste valuable memory (ASP.NET may flush cache to make room!) Never cache search results – disable search results layout page cache Never cache personalized web parts

Cache is but….

Setting memory based caching can waste valuable memory (ASP.NET may flush cache to make room!)

Never cache search results – disable search results layout page cache

Never cache personalized web parts

Optimizing for the WAN Page optimization techniques Use BLOB cache Use IIS Compression for dynamic pages Minimize images sizes and count Delay downloading of core.js Optimize list view pages (allitems.aspx) Use Fiddler ( http://www.fiddlertool.com ) Index Optimization Specialize crawl time schedules for WAN links Increase time-out settings for search Crawler impact rules/ # crawl threads

Page optimization techniques

Use BLOB cache

Use IIS Compression for dynamic pages

Minimize images sizes and count

Delay downloading of core.js

Optimize list view pages (allitems.aspx)

Use Fiddler ( http://www.fiddlertool.com )

Index Optimization

Specialize crawl time schedules for WAN links

Increase time-out settings for search

Crawler impact rules/ # crawl threads

Dev Techniques for the WAN Optimizing Web Parts for the WAN Re-use built-in Styles vs. Creating Custom Styles Storing State Use Server Cache class Web.config: enableViewState=true Use Control State Page.RegisterRequiresControlState() LoadControlState(), SaveControlState() Maximize perf on Webparts displaying data Use inline XML data islands Use Client-Side scripts to connect to Webservices AJAX and JSON is lightweight

Optimizing Web Parts for the WAN

Re-use built-in Styles vs. Creating Custom Styles

Storing State

Use Server Cache class

Web.config: enableViewState=true

Use Control State

Page.RegisterRequiresControlState()

LoadControlState(), SaveControlState()

Maximize perf on Webparts displaying data

Use inline XML data islands

Use Client-Side scripts to connect to Webservices

AJAX and JSON is lightweight

LAN Farms not WAN Conditions for supported configuration: WFE  1 millisecond(ms) latency  DB All servers on the same network segment Same VLAN No router switching SSPs in the same datacenter Servers cannot cross time zones

Conditions for supported configuration:

WFE  1 millisecond(ms) latency  DB

All servers on the same network segment

Same VLAN

No router switching

SSPs in the same datacenter

Servers cannot cross time zones

Agenda Geo Dilemma Deployment Options Information Architecture Search Architecture for the WAN Optimizing for the WAN WAN Performance Testing

Geo Dilemma

Deployment Options

Information Architecture

Search Architecture for the WAN

Optimizing for the WAN

WAN Performance Testing

WAN Test Hardware – MTC Austin SAN EMC CLARiiON CX700 30 x 73GBytes spindles @10,000RPM FC 10 spindles/LUN 3 LUNs, RAID 10 SQLServer Dell PowerEdge 6950 4 x Dual-Core AMD Opterons 2.60GHz 32 GB RAM SQL Server 2005 64-bit w/ SP2 WFE Dell PowerEdge 2950 2 x 3.0 GHz Dual-core Intel Xeon Processors 16 GB RAM MOSS 32-bit WAN Accelerator Shunra VE STN

SAN

EMC CLARiiON CX700

30 x 73GBytes spindles @10,000RPM FC

10 spindles/LUN

3 LUNs, RAID 10

SQLServer

Dell PowerEdge 6950

4 x Dual-Core AMD Opterons 2.60GHz

32 GB RAM

SQL Server 2005 64-bit w/ SP2

WFE

Dell PowerEdge 2950

2 x 3.0 GHz Dual-core Intel Xeon Processors

16 GB RAM

MOSS 32-bit

WAN Accelerator

Shunra VE STN

Test Topology

User Operations Test Mix Common User Operation Mix Description Percentage WSSHomePage Browse to the site’s homepage 25% WSSSmallDocLibAllItems Browse to all items view of a Document Library 14% WSSDocHTTPFetch Download a document via the browser 12% WSSGetListPageHtml View all items in a list 12% WSSSearch Search of pre-define search terms 11% WSSRSSList Read from an RSS list feed 10% WSSDispForm List Item view 9% WSSOutlookSync Perform a SOAP call to WSS to update a SharePoint List update in Outlook 7%

Page Load Times – Common User Operation Concurrency: Bandwidth (Mbps) 50 100 500 1000 10,000 Utilization (%) Throughput (Kbps) 45 (T3), Latency = 50ms 1.17 1.17 1.18 1.20 1.23 12 – 15 225 – 281 45 (T3), Latency = 100ms 1.87 1.90 1.90 1.90 1.90 7 – 9 134 – 174 45 (T3), Latency = 250ms 4.00 4.07 4.14 4.16 4.16 4 - 8 79 – 90 45 (T3), Latency = 500ms 7.50 7.50 7.70 7.90 7.90 2 – 4 35 - 48 12 (8 T1s), Latency = 50ms 1.31 1.34 1.35 1.36 1.37 43 – 48 231 - 260 12 (8 T1s), Latency = 100ms 2.03 2.03 2.04 2.07 2.08 30 – 32 156 - 180 12 (8 T1s), Latency = 250ms 4.14 4.20 4.25 4.26 4.35 13 – 15 75 – 86 12 (8 T1s), Latency = 500ms 7.60 7.70 7.80 7.80 7.90 6 – 9 40 - 50 6 (Quad T1), Latency = 50ms 1.93 1.94 1.94 1.97 1.88 69 - 72 184 - 203 6 (Quad T1), Latency = 100ms 2.27 2.34 2.37 2.38 2.40 53 - 60 144 - 162 6 (Quad T1), Latency = 250ms 4.20 4.21 4.28 4.35 4.46 22 – 32 62 - 90 6 (Quad T1) Latency = 500ms 7.70 7.90 7.90 8.10 8.10 15 – 18 43 - 50 3 (Dual T1), Latency = 50ms 3.40 3.43 3.55 3.61 3.65 65 - 90 90 - 124 3 (Dual T1), Latency = 100ms 3.57 3.59 3.68 3.69 3.72 64 – 85 88 - 116 3 (Dual T1), Latency = 250ms 4.51 4.57 4.68 4.71 4.79 44 - 60 61 - 82 3 (Dual T1), Latency = 500ms 7.70 8.00 8.00 8.10 8.20 27 – 35 40 - 50 1.5 (T1), Latency = 50ms 6.40 7.20 7.40 7.50 7.50 85 - 95 62 - 66 1.5 (T1), Latency = 100ms 6.50 7.10 7.20 7.20 7.60 72 - 86 52 - 60 1.5 (T1), Latency = 250ms 7.00 7.10 7.20 7.30 7.40 75 – 84 52 - 60 1.5 (T1), Latency = 500ms 8.40 8.50 8.70 8.80 8.90 60 – 65 42 - 47 0.5 (512K), Latency = 50ms 18.20 18.90 20.40 20.40 22.00 80 – 98 19 - 24 0.5 (512K), Latency = 100ms 18.30 18.90 20.40 20.40 22.00 80 – 96 18 - 24 0.5 (512K), Latency = 250ms 18.50 18.80 19.60 20.80 21.10 78 - 90 21 - 24 0.5 (512K), Latency = 500ms 18.80 19.00 20.00 21.90 22.70 73 - 92 17 - 22

Page Load Time: Common User Operation

Thank You!