Upload
ulmer
View
72
Download
0
Tags:
Embed Size (px)
DESCRIPTION
Microsoft Exchange Best Practices. Amy Styers – [email protected] Commercial Microsoft Solutions Consultant. What are “Best Practices”?. Best practices are accepted truths and wisdom based on: Manufacturer’s recommendations Historical evidence Analytical data Lessons learned - PowerPoint PPT Presentation
Citation preview
1EMC CONFIDENTIAL—INTERNAL USE ONLY
Microsoft Exchange Best Practices
Amy Styers – [email protected] Microsoft Solutions Consultant
2EMC CONFIDENTIAL—INTERNAL USE ONLY
What are “Best Practices”?
Best practices are accepted truths and wisdom based on:
– Manufacturer’s recommendations– Historical evidence– Analytical data– Lessons learned – Proof points
Best practices are general recommendations that:
– Provide guidance and considerations in the design stages
3EMC CONFIDENTIAL—INTERNAL USE ONLY
Best Practices are Based on The Audience
Depend on audience, requirements complexity, sophistication– Some we understand intuitively– Some we don’t. What do these represent?– Be flexible– What may be good for one implementation, it may not be good for another
F
4EMC CONFIDENTIAL—INTERNAL USE ONLY
Rule 1 - Understand Exchange I/O Profiles
Understand the various user profiles and how this information can be gained Also be aware of what else generates IO which needs to be considered in the
design
This is the foundation to properly sizing Exchange
5EMC CONFIDENTIAL—INTERNAL USE ONLY
Use the System Monitor tool to monitor Physical Disk\Disk Transfers/sec counter over the peak hours of server activity.
– This will allow for handling of random and “bursty” moments– Monday averaging between 8:30 or 11am is a typical peak average
Be aware of online maintenance activity. In some cases it can be very IO intensive and can generate as much activity as normal Exchange operations
IOPS Peak
Rule 1 - Understand Exchange I/O Profiles
6EMC CONFIDENTIAL—INTERNAL USE ONLY
Rule 2 – Size Exchange based on both I/O requirement and mailbox capacity
Old Way of Sizing (Just Capacity)– Number of Users x Mailbox Size x Growth = ~1234 GB– Buy enough disk to satisfy the capacity requirement (1234 GB /
disk size)– With larger disk sizes, there may not be enough spinning disks for
good performance
NEW Way of Sizing (Capacity and Performance)– Use enough physical disk spindles to satisfy total user IOPS and
total user mailbox capacity – Total IOPS required / Disk Spindle IOPS
Also know this – Be aware of how I/O read/write ratios in Exchange 2007 have
changed vs. Exchange 2003– Understand the tradeoff of large mailboxes– Apply as much detailed information to your design
7EMC CONFIDENTIAL—INTERNAL USE ONLY
Log write/s DB write/s DB read/s0
500
1000
1500
2000
2500
R/W ratios
Exchange 2003 4GBExchange 2007 4GBExchange 2007 8GBExchange 2007 12GBExchange 2007 16GBIO
PS
Rule 2 – Size Exchange based on both I/O requirement and mailbox capacity
Server memory configurations are key in Exchange 2007 design. Read/write ratios are typically 1:1 Be aware of your total server memory configuration requirements
8EMC CONFIDENTIAL—INTERNAL USE ONLY
0
20
40
60
80
100
120
140
160
180
0 100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400 1500 1600 1700 1800 1900 2000
Mailbox Size (MB)
Driv
es
Required for Capacity
ExcessCapacity
Exchange 2003Exchange 2007
146 GB 10KRAID 5
300 GB 10KRAID 1/0
146 GB 15KRAID 1/0
Required for IOPS
Paying for moreI/O than needed?
Required for IOPS
Rule 2 – Size Exchange based on both I/O requirement and mailbox capacity
10EMC CONFIDENTIAL—INTERNAL USE ONLY
IOPS: the number of Input Output operations (I/O’s) per second
%R: Percentage of I/O’s that are Reads WP: RAID Write Penalty (RAID 1 = 2, RAID 5 =
4) %W: Percentage of I/O’s that are Writes Factor other variable such as archiving, journaling,
virus protection, remote devices, and risk
Don’t solely rely on automated tools in sizing your Exchange environment. Place detailed
effort into your calculations and provide supporting factual evidence on your
designs rather than providing fictional calculations
Rule 2 – Size Exchange based on both I/O requirement and mailbox capacity
11EMC CONFIDENTIAL—INTERNAL USE ONLY
Rule 3- RAID Protection Type
With Exchange 2007, technically any RAID type can satisfy the IO, as long as there are enough spindles to deal with the increased write ratio. However consider the following
– Understand what each type provide – Performance, reliability– Cost, practicality– Risk, exposure
RAID 1 Best practice– Always works, predictable and best
performance
RAID 5 & RAID 6– Writes do impact reads– Rebuilds impact reads and writes
12EMC CONFIDENTIAL—INTERNAL USE ONLY
Rule 4- Spindle and Storage Best Practices
Avoid sharing physical spindles over multiple servers– Best: an Exchange server has its own physical spindles, i.e. the
building block approach– Possible: Exchange servers share physical spindles– Possible, with care: Exchange servers share physical spindles with
BCV volumes used for their own backup– Possible, but not good: Exchange servers share physical spindles
with other applications with predictable, known, and similar workloads.
– Very bad: Exchange servers share physical spindles with applications with unpredictable and incompatible workloads, like data warehouses.
– If unable to dedicate spindles to Exchange, be aware of what else is running on the physical spindle, and account for that activity
13EMC CONFIDENTIAL—INTERNAL USE ONLY
Performance problems may arise from sharing physical disk resources, with other I/O intensive applications and databases.
Exchange database write characteristics are very random and small in size
Avoid sharing Exchange physical disks with other application like different IO characteristics like Oracle
These recommendations are true for both DAS and SAN topologies.
Understand the storage technology options available your Exchange environment
Understand the business requirements first
Not dedicating physical disk to exchange can bring
unpredictable performance
Dedicating physical disk to exchange allows you to
maintain predictable levels of performance
Rule 4- Spindle and Storage Best Practices
14EMC CONFIDENTIAL—INTERNAL USE ONLY
Exchange on direct attached storage (DAS/JBOD)– Exchange always ran on direct attached storage, but…
Reliability issues Lack of write cache No rebuild priorities No remote replication, while Exchange was becoming business critical DAS and virtualization don’t mix well – Virtualization is a key market trend
Enterprise Storage Options – Messaging environments are mission critical applications
Treat them as such – Understand your technology options– Understand your technology cost
Acquisition cost Long term of ownership ( all related cost )
Rule 4- Spindle and Storage Best Practices
15EMC CONFIDENTIAL—INTERNAL USE ONLY
Rule 5 – Design Methodology for Exchange
Iterate over all possible solutions, until the right solution has been found
Platform independent Considers total solution, including
software Parameters of an Exchange
configuration determine the possibilities
List accuracy of possibility Be aware of Exchange object
requirements such as ESG configuration requirements
Exchange should be properly planned before disks are laid out. EMC recommends the use of Building Blocks outlined in our solutions and ESRP, this makes the design,
install, backup & troubleshooting much predictable
16EMC CONFIDENTIAL—INTERNAL USE ONLY
Myths Around Sharing Logs and Data on Same Physicals Microsoft does not allow it! FALSE
– OK when you can recover from a double RAID 1 device failure– http://www.microsoft.com/technet/prodtechnol/exchange/guides/StoragePerformance/fa839f7d-f8
76-42c4-a335-338a1eb04d89.mspx
Separate logs and data because of performance, it is TRUE for solutions without write cache
– Sometimes write cache is lost after a board failure, requiring separation of logs and data volumes also
Rule 5 – Design Methodology for Exchange
17EMC CONFIDENTIAL—INTERNAL USE ONLY
Share logs and data on physical spindles (Symmetrix)– Better usage of available spindles– Do NOT share logs and data on same logical volume (META)– Logs and DB for the same SG should not reside on the same spindles
Subsystem Cache Will Not Guarantee Performance– The quantity or intelligence of cache does not override normal disk sizing and layout
advice– Provide enough spindles
Optimizer is not the magical spell either– Is not the magic that will automatically "create" a good configuration over time– Exchange, being such a random application, should be planned out properly before
disks are laid out
Rule 5 – Design Methodology for Exchange
18EMC CONFIDENTIAL—INTERNAL USE ONLY
Use Striped Metavolumes / MetaLUNs– Much better than concatenated in performance tests
Symmetrix (Metavolumes)– Multiples of four and handled well by Symm Disk Adapters– Example: 4 member meta, 8 member meta– Striped metavolumes need a BCV to expand
CLARiiON (MetaLUNs)– Great performance, simple to expand
Host Volume Sets Require Windows Dynamic Disks– Present limitations in clusters, replication– Stay with Basic Disk
(2) RAID10 (8+8)’s to make MetaLUN
+ + + + + +
Rule 5 – Design Methodology for Exchange
19EMC CONFIDENTIAL—INTERNAL USE ONLY
Server– Align using 64 KB offset with Diskpart ( Widows 2003 )– Set allocation unit size on DB and Log LUNs to 64 KB– Set HBA QDepth to 128– EMC PowerPath for load balancing
Array– Keep default 8 KB page size on CLARiiON– 8 KB sector within new DMX3 cache slot provides better fit– Don’t change LUN offsets
HBA and FA ports– Fan-out two HBA to at least four FA ports– FA gives highest priority to queuing I/O requests from host– During a write burst a FA can be overloaded– More FA avoids overload– It is NOT needed to dedicate four FA ports– Share FA ports with other Exchange servers– Do NOT share Exchange with high bandwidth servers on the same FA slices/CPU– Increase queue depth to 128 ( or higher when possible )– Server has set HBA queue depth, and HBA has to set the queue depth to 128
Use PowerPath to:– Give priority to log writes– Throttle writes to avoid starvation of reads
Rule 6 – Consider All Elements in the Design
20EMC CONFIDENTIAL—INTERNAL USE ONLY
Understand Bottlenecks/Weakest Link
Application (users, layout, tuning, AD, isolation) Volume Manager (format, alignment, stripe
size) Host Bus (HBA queue depth, PowerPath) Disk (spindles, metavolumes, replication,
archiving)
Rule 6 – Consider All Elements in the Design
21EMC CONFIDENTIAL—INTERNAL USE ONLY
Rule 7: Test and Measuring Performance
Besides EMC Workload Analyzer/Performance Manager and NaviAnalyzer use the following tools:
ExBPA – Understand where you are today, use a base line
information. JetStress
– Suggested by Microsoft for design testing – simulates I/O work load
LoadGen– Simulates email load based on profiles
Perfmon– Help you understand current I/O characteristics
TEST !!! Never go live without proper testing
22EMC CONFIDENTIAL—INTERNAL USE ONLY
JetStress generates the same throughput as Exchange in a production environment
– Each thread generates a mix of transactions (insert, delete, replace and seek) to mimic an Exchange workload
– Increasing thread count increases the throughput to match what is expected in production
Mailbox count, and I/O profile are used to calculate expected IO/s
– Is only used to determine when a test has exceeded the expected IO/s and rated “successful”
– There is NO difference between 50,000 users at 0.1 Expected IOPS 5,000 5,000 users at 1.0 Expected IOPS 5,000
– There is a difference in real life!
Rule 7: Test and Measuring Performance
23EMC CONFIDENTIAL—INTERNAL USE ONLY
Use JetStress to verify the performance and stability– JetStress is used to validate the performance of a disk subsystem prior to putting an
Exchange server into production. – JetStress helps verify disk performance by simulating Exchange – Watch out for very high user counts based upon JetStress– Mailbox size is used to create the initial databases– The stroking distance is always 100%
Always do JetStress before deployment– Any mistake, any imbalance will be made visible with the opportunity to correct before
implementing in production– Compare against EMC ESRP results– Watch out for caching effect
Test with LoadGen after JetStress testing – Microsoft Exchange Load Generator (LoadGen) is a simulation tool to measure the
impact of MAPI, OWA, IMAP, POP and SMTP clients on Exchange servers – LoadGen can require a lot of hardware, but is able to reproduce results
Rule 7: Test and Measuring Performance
24EMC CONFIDENTIAL—INTERNAL USE ONLY
PhysicalDisk\Average Disk sec/ReadIndicates the average time (in seconds) to read data from the disk.
The average value should be below 16 ms.Spikes (maximum values) should not be higher than 20 ms.
PhysicalDisk\Average Disk sec/WriteIndicates the average time (in seconds) to write data to the disk.
The average value should be below 16 ms.Spikes (maximum values) should not be higher than 20 ms.
PhysicalDisk\Average Disk Queue LengthIndicates the average number of both read and write requests that were queued for the selected disk during the sample interval.
The average should be less than the number of spindles of the disk. If a SAN is being used, ignore this counter and concentrate on the latency counters: PhysicalDisk\Average Disk sec/Read and PhysicalDisk\Average Disk sec/Write.
Do validate latest acceptable latency limits
Rule 7: Test and Measuring Performance
Understand what the latency acceptable limits are:
25EMC CONFIDENTIAL—INTERNAL USE ONLY
Do validate latest acceptable latency limits
Rule 7: Test and Measuring Performance
Understand what the latency acceptable limits are:
26EMC CONFIDENTIAL—INTERNAL USE ONLY
Rule 8: Help Plan the Entire Architecture
Storage is usually an afterthought Getting into Exchange and Active Directory Planning sessions will help you
understand the environment better Poorly Planned Active Directory = Exchange Performance Problems may manifest EMC offers services that can evaluate the quality of a customers Active Directory
deployment
– Exchange Insight Workshop– Migration Assessment– Migration Design– Migration Implementation
Tie it all together– Business requirements– Storage Sizing– Performance– Back up – Restore– Recovery– Distance Replication– Security– Management – Archiving
27EMC CONFIDENTIAL—INTERNAL USE ONLY
Rule 9 - Backup Best Practices
Avoid unprotected clones and mirrored clones with VSS– Recovery after a drive failure is VERY difficult– Same when drive fails while mirroring from M1BCV to M2BCV
RAID 5 clones are the best solution in combination with VSS– Recovery after drive failure is simple– Backup/restore operation can continue with a bad drive, but slow
Protected clones are possible too– Uses two mirror positions
SNAPs are possible, but consider the following – Change rate– Activity during ESEUTIL, on-line maintenance and backup– Sequential read is not optimal – long term consequences– Understand the difference between recovery and restore
Backup and restore have same granularity with hardware VSS
28EMC CONFIDENTIAL—INTERNAL USE ONLY
Complicated management– 50 SG have 100 LUNs to manage– Trend to have fewer SGs ( 8 to 16 )
Map backup order over time– Avoid contention by carefully mapping out backup and eseutil processes.
Very hard to do. More SGs make it more complicated One sequential stream per clone 1:1 map from source to clone
Concurrent backups– Have as many backup threads as possible to get to the required throughput.
Sequential with random seeks to the clone drive Still fast enough to meet the windows. Uses a big pool of clone devices
Make sure you understand the importance of backups– Don’t just recommend “no backups” due to technology limitations– Backups are subject to regulatory requirements
Rule 9 - Backup Best Practices
29EMC CONFIDENTIAL—INTERNAL USE ONLY
Break data into atom (sub-file, variable-length segments of data)
Send and store each atom only once
Avamar backup repository
…up to 500 times daily data reduction
At the source—De-duplication before data is transported across the networkAt the target—Assures coordinated de-duplication across sites, servers, and over timeGranular—Small, variable-length segments guarantee most effective de-duplication
Include data de-duplication as part of the backup strategy
Take advantage of VSS technology and off-load the backup mechanism off the production servers
O H
OH H
OH H
OH H
OH H
OH H
Rule 9 - Backup Best Practices
30EMC CONFIDENTIAL—INTERNAL USE ONLY
Rule 10 – Leverage Server Virtualization Technology
ResultServer hardware
trends driving Exchange
virtualization
ResultESX 3.5 is ready
for Exchange 2007
Server hardware changes- Exchange 2007 requires 64-bit servers- Now ship with multi-core 2/4/(soon 6) CPUs, 256GB RAM- Intel/AMD hardware-assisted virtualization - Huge potential for under-utilization- Opportunity to consolidate and reduce costs
Changes in ESX 3.5 and beyond- Increased guest OS memory (64GB)- Increased physical RAM on ESX (256GB)- Network improvements lower CPU utilization- NUMA optimizations improve multiple VM performance- Improved storage efficiency
Server Virtualization technology is a great fit for Exchange- Position VMware whenever possible - Provides great TCO to Exchange deployments – much better than DAS- Understand the latest testing on virtualization technology
31EMC CONFIDENTIAL—INTERNAL USE ONLY
Virtualizing Microsoft Applications • Historically, some customers feared virtualization of enterprise
applications.• High utilization of resources limited by 32-bit system• Large amount of IO traffic • Server memory limitations
• But, hardware has changed…• From dual-core to multi-core technology increase available
resources• More memory per server
• Applications like Exchange have changed…• better uses multiple cores, 64-bit• Reduced I/O• more server roles to consider
• Microsoft Application are excellent candidates for virtualization!• Benefits of virtualization, such as encapsulation/portability,
hardware availability, and change control are now accessible in the world of Microsoft applications
Rule 10 – Leverage Server Virtualization Technology
32EMC CONFIDENTIAL—INTERNAL USE ONLY
Dispelling Myths - Performance
• Many still convinced Exchange should never be virtualized• Exchange 2007 performance testing proves otherwise• What performance is important to Exchange administrators?
- Low latency within defined thresholds
- Exceptional user experience
- Available headroom for future growth
- Constant latency while scaling to multiple, concurrent mailbox servers
- Minimal storage latencies for large mailbox counts
- Flexibility to meet these requirements under changing workloads
Let’s take a look !!!
Rule 10 – Leverage Server Virtualization Technology
33EMC CONFIDENTIAL—INTERNAL USE ONLY
• 16,000 “heavy” users on single server • CLARiiON CX3-80• Replication Manager• Dell R900
- 16 core- 128GB RAM
• Virtual machines:- 4,000 users each- 4 vCPU- 12GB RAM
• Whitepaper:
Rule 10 – Leverage Server Virtualization Technology
34EMC CONFIDENTIAL—INTERNAL USE ONLY
Leverage Dynamic Adaptability with Virtual LUN Technology
Unanticipated virtual machine growth New performance needs changed Alter performance of the virtual
machine file system without disruption– Move between disk or RAID types
Non-disruptive to virtual machines and applications
– Adjust for unplanned changes Tier virtual machine groups
– Efficient usage of storage Worry-free adaptability
Virtual Machines
APP
OS
APP
OS
APP
OS
APP
OS
APP
OS
APP
OS
APP
OS
APP
OS
VMware ESX Servers
APP
OS
APP
OS
APP
OSSATA II SATA II SATA II SATA II SATA II SATA II
FC FC FC FC FC FCFC FC FC
Virtual LUN Technology
Rule 10 – Leverage Server Virtualization Technology
35EMC CONFIDENTIAL—INTERNAL USE ONLY
Rule 10 – Leverage Server Virtualization Technology
In general Exchange is not a good candidate for thin LUN’s
– Tier 1 application– Heavy, bursty IO activity– Latency intolerant– Preference for RAID 1/0– Often reaches max storage quickly
But some Exchange installations may qualify
– Smaller user counts– Low IO’s per user– Mailboxes that won’t reach their max for
more than a year
VP configuration– Dedicated Storage Pool– Periodic provisioning– NQM cruise control with mixed use
Saves management and capital costsMailbox size
IO’s
per
use
r
Small
Low
Large
High
VirtualProvisioningCandidates
Provided by CX partner engineering
36EMC CONFIDENTIAL—INTERNAL USE ONLY
High Priority Medium Priority
LowPriority
Avai
labl
e Pe
rfor
man
ceApplications
Maximize Resources - QoS Manager and DRS
Concerned about introducing contention with consolidation?
DRS balances host resources– Monitors CPU & memory utilization– Leverages policy-based VMotion
NQM enforces application storage performance policies
– Throughput, Bandwidth, Response Together offer end-to-end policy-based
performance management
This is advance functionality you will not find in JBOD or DAS
VMware ESX Server
APP
OS
APP
OS
APP
OS
APP
OS
APP
OS
APP
OS
Rule 10 – Leverage Server Virtualization Technology
38EMC CONFIDENTIAL—INTERNAL USE ONLY
CORE STORAGE BEST PRACTICES
39EMC CONFIDENTIAL—INTERNAL USE ONLY
Core Storage
1. Align all Microsoft Exchange-related disks using a value of 64.
– This aligns all of the Exchange-related NTFS partitions on a 64-KB boundary. With the release of Windows 2008, this issue has been addressed and corrected, so it is no longer necessary to perform this task in disk configured in Windows 2008.
40EMC CONFIDENTIAL—INTERNAL USE ONLY
Core Storage
2. Format all Microsoft Exchange-related NTFS partitions using 64-KB Allocation Unit (AU) cluster size.
– While this cluster size has been shown to have no effect on normal Microsoft Exchange database operations (transaction activity), studies have shown that a 64-KB cluster size increases performance with certain Microsoft Exchange and NTFS-related operations, such as Exchange backups and Exchange check-summing activities associated with VSS-related operations.
41EMC CONFIDENTIAL—INTERNAL USE ONLY
Core Storage
3. Isolate the Microsoft Exchange database workload from other I/O-intensive applications or workloads.
– This ensures the highest levels of performance for Microsoft Exchange and makes troubleshooting easier in the event of a disk-related Microsoft Exchange performance problem.
42EMC CONFIDENTIAL—INTERNAL USE ONLY
Core Storage
4. Separate logs and databases onto different disks and RAID groups. – This may pose performance issues as database, and log files have very different I/O
characteristics. It can also be an issue if placing log files and databases from the same ESG in a given volume group, as certain recovery options can be impaired. If desired, it is possible to combine logs, and databases of separate Exchange storage groups within the same physical spindle (typically in the Symmetrix.) Do NOT share logs and data on same logical volume (META). Microsoft has acknowledged this recommendation, and has updated this KB article.
43EMC CONFIDENTIAL—INTERNAL USE ONLY
Core Storage
5. Size and configure the environment for spindle performance as a primary consideration, with spindle or storage capacity as a secondary concern. In other words, size for performance first and then capacity requirements and performance results
Microsoft has released a new Exchange 2010 sizing tool
44EMC CONFIDENTIAL—INTERNAL USE ONLY
Core Storage
6. Tuning the array storage system parameters is important in obtaining best performance. The following list details the optimal parameters for Exchange:
Cache page size of 8 KB Maximized write cache size Read and write cache enabled for all LUNs
45EMC CONFIDENTIAL—INTERNAL USE ONLY
Core Storage
6. Use a Building Block approach for planning storage for Exchange