Upload
winwire-technologies-inc
View
3.344
Download
0
Embed Size (px)
Citation preview
© 2010 WinWire TechnologiesWinWire Technologies, Inc. Confidential
Planning Very Large Scale Document Repositories with High Availability (HA) in
SharePoint Server 2013
@WinWire
Wired2Win Webinar Series
WinWire Technologies Inc.2350 Mission College Boulevard,Suite 480, Santa Clara, California 95054 Phone/Fax: +1 (888) 849 7339 Email : [email protected]
© 2010 WinWire TechnologiesWinWire Technologies, Inc. Confidential2
© 2010 WinWire TechnologiesWinWire Technologies, Inc. Confidential3
Who We Are
WinWire Technologies is a specialized IT solutions
company focused on helping business and technology
leaders achieve an “on-the-move” business environment
leveraging pre-built collaborative and analytic solution accelerators and
cross-platform mobile technologies.
© 2010 WinWire TechnologiesWinWire Technologies, Inc. Confidential4
This is the story of Bob.
© 2010 WinWire TechnologiesWinWire Technologies, Inc. Confidential5
Why is Bob sad?
© 2010 WinWire TechnologiesWinWire Technologies, Inc. Confidential6
Bob deployed SharePoint.
© 2010 WinWire TechnologiesWinWire Technologies, Inc. Confidential7
Things were pretty good.
© 2010 WinWire TechnologiesWinWire Technologies, Inc. Confidential8
So what’s the problem?
© 2010 WinWire TechnologiesWinWire Technologies, Inc. Confidential9
He needs to provide the following functionality to his organization.
Very Large Scale Document
Repositories
Highest possible availability
Predictable “Elastic”
architecture that can grow with the
deployment
Lowest possible maintenance and operations costs
© 2010 WinWire TechnologiesWinWire Technologies, Inc. Confidential10
He’s got many folks to please, including IT Support, legal and compliance.
© 2010 WinWire TechnologiesWinWire Technologies, Inc. Confidential11
Users are concerned.They don’t want it to be hard.
© 2010 WinWire TechnologiesWinWire Technologies, Inc. Confidential12
SharePoint seems to provide the right capabilities!
© 2010 WinWire TechnologiesWinWire Technologies, Inc. Confidential13
Or does it?
© 2010 WinWire TechnologiesWinWire Technologies, Inc. Confidential14
Bob isn’t sure how to proceed.
© 2010 WinWire TechnologiesWinWire Technologies, Inc. Confidential15
Let’s help Bob withPlanning Very Large Scale Document Repositories with High Availability (HA) in SharePoint
© 2010 WinWire TechnologiesWinWire Technologies, Inc. Confidential16
Agenda
© 2010 WinWire TechnologiesWinWire Technologies, Inc. Confidential17
Agenda
© 2010 WinWire TechnologiesWinWire Technologies, Inc. Confidential18
Enterprise Content Management
Create Control Protect
Create and organize content easily
Manage content policy, information architecture and taxonomy
Reduce risk and manage compliance with centralized tools
© 2010 WinWire TechnologiesWinWire Technologies, Inc. Confidential19
• Motivation• End-user experience is what really
matters.• Tail end of performance is sticky.
• Basic Idea• Categorize applications into SLA
buckets.
• Homogenize front-end, backend, and data layers.
• Prevent proliferation of app server deployments.
SharePoint 2013 Designing Your Farm
© 2010 WinWire TechnologiesWinWire Technologies, Inc. Confidential20
Designing for better SLA
Latency Throughput Resource Utilization
DesiredBehavior
Very low < 5 msec
Very High MediumConsistent Latency
Low <500 msec
Medium Low-MediumConsistent Latency
High> 1 min
HighHigh-VeryHigh
High Throughput
Very Low < 5 msec
Very High Low-MediumConsistent Latency
Front-End
Batch Processing Layer
Specialized Workload
Search
Database Layer
Request Management
Distributed Cache
Latency Throughput Resource Utilization
DesiredBehavior
Very low < 5 msec
Very High MediumConsistent Latency
Low <500 msec
Medium Low-MediumConsistent Latency
Low <500 msec
Medium Low – HighFairly Consistent Latency
High> 1 min
HighHigh-VeryHigh
High Throughput
Very Low < 5 msec
Very High Low-MediumConsistent Latency
© 2010 WinWire TechnologiesWinWire Technologies, Inc. Confidential21
< 10K Users
More Workloads
More Resources
< 10M Items
<100 Users
Evaluation
Very Light And Simple Workload
< 1,000 Users
Simple Workload
Small Content
Fault Tolerance
More & More Users
More Workloads
More Usage
Large Content
Multiple Index Partitions
Advanced Routing
Routing &
Caching
Front-End
Batch
Processing
Database
Crawl
Query
Search Specialized
Schematic Diagram…Machines could be virtual or physical.Topologies are rough estimates.
Evolution of topology
© 2010 WinWire TechnologiesWinWire Technologies, Inc. Confidential22
WORKLOAD
Facts: purpose, services, # of users, concurrency, rps, location
Tuning options: satellite farms, disable services, block clients …
Facts: #of site collections, DBs, data size
Tuning Options: split site collections, balance content…
Facts: # of servers, hw spec, roles, network
Tuning Options: add server, scale-up, tune settings …
What defines your SharePoint?
© 2010 WinWire TechnologiesWinWire Technologies, Inc. Confidential23
Microsoft’s 2013 Dogfood Farm
WORKLOAD
DATASET
HARDWARE
SLAs (During Peak Hours)1.3 TB total dataset size
1,001,141 documents
10 content DBs
Largest content DB - 290 GB
8,297 Site collections
Largest site collection - 275 GB
15,000 users
8,800 active users per week
1.7M requests per day
Collab, Social, Doc Mgmt, Project
204,106 profiles in UPA
1 Web App
Latency(99th ntile)
511 ms
RPS
50
Active Users (hr)
2,500
Availability
99.9%
Role HW Spec
Routing & Caching VM, 4 cores, 14GB RAM
Front End VM, 4 cores, 14GB RAM
Batch Processing VM, 4 cores, 14GB RAM
Database 8 cores, 64 GB RAM
Role Avg CPU/ Mem
Routing & Caching 12% / 8 GB
Front End 45% / 11GB
Batch Processing 80% / 12 GB
Database 11% / 46 GB
© 2010 WinWire TechnologiesWinWire Technologies, Inc. Confidential24
Capacity Planning Resources
© 2010 WinWire TechnologiesWinWire Technologies, Inc. Confidential25
Component CPU Network Disk Memory
Search administration 1
Crawler 2
Content processing (CPC)
3
Analytics processing (APC)
3
4
Index
Query processing (QPC) 5
1. Admin: Memory footprint increase with topology size
2. Crawler: Crawled documents are temporarily stored to
disk
3. CPC and APC are scheduled with “below normal”
priority
4. APC: Network bandwidth increase with scale out, but
only in-between APC nodes
5. QPC: Network IOPS increases linearly with number of
index partitions and QPS
Components – Scaling Cheat sheet
© 2010 WinWire TechnologiesWinWire Technologies, Inc. Confidential26
Sample Worksheet
Current Size Recommended storage
Notes
Configuration 2 GB The configuration database grows very slowly. Approximate growth is 40 MB per 50,000 Site Collections
Central Administration Content
1 GB Very slow growth. Not expected to grow beyond 1 GB.
Content (one or more) ????????????
• Functional Area (PROD) • Portal and Self Service• Legacy Data to be migrated. When migration of legacy data is performed, database sizing and
Site Collection design has to take place.• Cross Functional Area Collaboration
Usage and Health Data 10 GB Dependent on logging level selected
Subscription Settings Service 5 GB Dependent on number of alerts created by users
Search Administration 10 GB Search Administration is typically small
Crawl (one or more) 50 GB Crawl = 0.046 x (sum of content databases)
Property (one or more) 15 GB Property = 0.015 x (sum of content databases)
Profile 10 GB Approximately 1 MB per user profile
Synchronization 5 GB Approximately 650 kb per user profile
Social Tagging 2 GB Approximately 0.009 MB per tag
Staging 5 GB Web Analytics Service
Reporting 5 GB Web Analytics Service
Secure Store 1 GB
Word Automation Services 1 GB
Managed Metadata 10 GB The size of the database is affected by the number of content types and keywords.
Total ???
© 2010 WinWire TechnologiesWinWire Technologies, Inc. Confidential27
Agenda
© 2010 WinWire TechnologiesWinWire Technologies, Inc. Confidential28
Results from SharePoint 2013 Deployments
• 50% faster server response
• 4 X faster profile sync. • 10 X faster w/ Active Directory import
• 40% more efficient use of bandwidth
• 80% less file i/o related SQL IOPS
• More scalable SharePoint farms
© 2010 WinWire TechnologiesWinWire Technologies, Inc. Confidential29
SharePoint Farms Scale More than Ever
Product does have its limits
Limits are for each
Dimension
Each dimension needs to
stay within limits
© 2010 WinWire TechnologiesWinWire Technologies, Inc. Confidential30
SharePoint 2013 Performance Benchmarks
SharePoint 2010 benchmark on SharePoint 2013
SharePoint 2013
SharePoint 2010
4 front-end VMs w/ 4 CPUs, 12 GB RAM Green Zone: < 60% CPU on FEs, < 0.5 sec server response time
1 physical SQL Server w/ 8 CPUs, 32 GB RAM Max Zone: 90+% CPU on FEs, < 1 sec server response time
© 2010 WinWire TechnologiesWinWire Technologies, Inc. Confidential31
SharePoint 2013 Scale Divisional Benchmark
Collaboration + social + communities + meta-data navigation
Green Zone: < 60% CPU on FEs, < 0.5 sec server response time
Max Zone: 90+% CPU on FEs, < 1 sec server response time
© 2010 WinWire TechnologiesWinWire Technologies, Inc. Confidential32
Agenda
© 2010 WinWire TechnologiesWinWire Technologies, Inc. Confidential33
SharePoint Security Model: Conceptual
Token
TOKENUser
GroupsRoles
Claims
AuthenticationProvider
Identities
STSUserDS
AuthenticationMethod
Authorization
User
GroupRole
AssignmentPermission
Level
F D
RC
Security Scope
Site
Policies
ANONYMOUSACCESS & POLICY
USER POLICY
Web App
PermissionsGroupDS
Identity Authentication
© 2010 WinWire TechnologiesWinWire Technologies, Inc. Confidential
End User Security Recommendations - Inheritance
34
• Roles assignments are inherited by child objects• Subsite, list, library, folder, item or document
• Best Practice• Use inheritance wherever possible
Role Assignment
Role Assignment
Site
W
Library
© 2010 WinWire TechnologiesWinWire Technologies, Inc. Confidential
End User Security Recommendations - Inheritance
35
● Roles assignments are inherited by child objects– Subsite, list, library, folder, item or document
Role Assignment
Role Assignment
Site
W
Library
Scope 5
AD Group X (Reader)
User 3 (Contributor)
User 4 (Full Control)
Scope 3
User 1 (Contributor)
Scope 4
User 2 (Contributor)
Web Object
Document Library Object
Folder Object
Item 1 Object
Item 2 Object
Item 3 Object
Scope 2
User 5 (Reader)
+ User 1 (Limited Access)
+ User 2 (Limited Access)
1
1
1
1
1
1
SPGroup Object FullGP
+ User 3
+ User 4
SPGroup Object ContributeGP
+ User 1
+ User 2
SPGroup Object ReaderGP
+ User 5
+ AD Group X
Scope 1
+ FullControlGP (Full Control)
+ ContributorGP (Contributor)
+ ReaderGP (Reader)
© 2010 WinWire TechnologiesWinWire Technologies, Inc. Confidential36
You can break inheritance– Assign explicit permissions– Dynamic security changing code redesign– Recommendation is to modify scope architecture so
that scope membership does not cause ACL recalculation at the parent document library and Web. Only possible using custom code.
Role Assignment
Role Assignment
Site
W
Library
X
End User Security Recommendations - Fine-grained Permissions
© 2010 WinWire TechnologiesWinWire Technologies, Inc. Confidential
End User Security Recommendations - Fine-grained Permissions
37
Role Assignment
Role Assignment
Site
W
Library
X
Scope 2
User 5 (Reader)
Scope 5
AD Group X (Reader)
User 3 (Contributor)
User 4 (Full Control)
Scope 3
User 1 (Contributor)
Scope 4
User 2 (Contributor)
Scope 1
User 2 (Reader)
User 3 (Full Control)
User 6 (Contributor)Web Object
Document Library Object
Folder Object
Item 1 Object
Item 2 Object
Item 3 Object
+ AccessGP1 (Limited Access)
1
2
3
4
1
5
SPGroup Object AccessGP1
+ AD Group X
+ User 3
+ User 4
+ User 5
+ User 1
+ User 2
+ User 1 (Limited Access)
+ User 2 (Limited Access)
© 2010 WinWire TechnologiesWinWire Technologies, Inc. Confidential
Agenda
38
© 2010 WinWire TechnologiesWinWire Technologies, Inc. Confidential
Production Farm Disaster Recovery Architecture
39
© 2010 WinWire TechnologiesWinWire Technologies, Inc. Confidential40
Disaster Recovery Options
Standby Type
Definition Description Pros Cons
Hot Standby A second data center that can provide availability within seconds or minutes.
Multiple data centers
Content and services through only one data center
Relatively fast to recover Can be quite expensive to configure and maintain
Warm Standby
A second data center that can provide availability within minutes or hours
Virtual server images shipped to local and regional disaster recovery farms
Relatively inexpensive to recover, because a virtual server farm can require little configuration upon recovery
Can be very expensive and time consuming to maintain
Cold Standby
A second data center that can provide availability within hours or days
Backups are shipped to support bare metal recovery to local and regional offsite storage on a regular basis, and contracts are in place for emergency server rentals in another region
Often the cheapest option to maintain, operationally
Often an expensive option to recover, because it requires that virtual servers be configured correctly after a disaster has occurred
The slowest option to recover
© 2010 WinWire TechnologiesWinWire Technologies, Inc. Confidential41
Agenda
© 2010 WinWire TechnologiesWinWire Technologies, Inc. Confidential
High availability (HA) with RBS
42
Size (GB)
ReductionWithout RBS
With RBS
Database size (100 GB) 217.2 7.0 96.8%
Database data file size
(100 GB) 106.9 3.2 97.0%
Database transaction log
file size (100 GB) 111.6 3.8 96.6%
Size of RBS externalized
data -- 96.2 --
Database size (1 TB) 2,292 26 98.9%
Database data file size (1
TB) 1,120 6.5 99.4%
Database transaction log
file size (1 TB) 1,173 20 98.3%
Size of RBS externalized
data -- 1,115 --
Effects of RBS on SQL Server Database Size
© 2010 WinWire TechnologiesWinWire Technologies, Inc. ConfidentialWinWire Technologies, Inc. Confidential
Topic: Line of Business (LOB) + Enterprise Search
When: December 05, 2013
Time: 8AM PST
Presented by: Bill Baer and Cem Aykan - Microsoft
Q & A
Next Webinar
© 2010 WinWire TechnologiesWinWire Technologies, Inc. ConfidentialWinWire Technologies, Inc. Confidential
Reach out to us at [email protected]
Visit us: www.winwire.com
http://www.winwire.com/winwire-blog/
www.twitter.com/winwire
Thank You