Vce Exchange Solution Architecture

8/2/2019 Vce Exchange Solution Architecture

1/27

2011 VCE Company, LLC. All rights reserved. 1

VBLOCK SOLUTION FOR MICROSOFTEXCHANGE 2010

April 2011


2/27


Table of Contents

Executive Summary .............................................................................................. 3Scope ................................................................................................................ 3Objectives ......................................................................................................... 4Audience ........................................................................................................... 4Terminology....................................................................................................... 4

Solution Architecture ............................................................................................. 5Vblock Platform Overview ................................................................................. 5Solution Components ........................................................................................ 6

Solution Validation .............................................................................................. 14Overview ......................................................................................................... 14Summary ......................................................................................................... 15

Conclusion .......................................................................................................... 16References ......................................................................................................... 18Appendix A JetStress 2010 Results ................................................................. 19Appendix B Load Generator 2010 Results ....................................................... 23Appendix C Best Practices .............................................................................. 26


3/27


Executive Summary

This document describes a Solution Architecture for designing and scaling Microsoft Exchange Server2010 with SP1 (hereinafter, Exchange 2010) on Vblock Infrastructure Platforms. Specifically, thissolution is built on the Vblock Series 700 model MX.

As enterprises migrate to virtualized environments for Exchange 2010 mailbox administration, a newsolution is required to manage larger mailbox sizes and number of users, while maintaining low per-

user mailbox costs. Exchange 2010 virtualization and consolidation significantly reduces cost, power,and footprint, while providing application mobility and agility to add more resources as needed.

By using the Vblock Infrastructure Platforms with Exchange Server 2010, larger numbers of Exchange

2010 IOPS can be produced in comparison to Direct Attached Storage configurations, resulting infewer spindles needed to produce the desired performance.

The Vblock Solution for Exchange 2010 provides an efficient path to meet the increased demandsof knowledge workers while maximizing performance for end users, and balancing the storage and

compute resources.

The Solution Architecture design approach and provisioning strategy for Exchange 2010 described inthis paper outlines how Vblock Platforms can scale the messaging infrastructure to provide a largenumber of users (for example, 20,000 users) with greater mailbox size and features, while maintaining

the lowest possible cost per mailbox.

The solution architecture described is based on a building block approach for designing and operatingExchange 2010. The building block approach is a scaling model for Exchange 2010 roles and their

required compute and storage resources. An organization can leverage the Vblock UnifiedInfrastructure Management (UIM) to provision UCS blade servers, VMware virtual servers, and to

orchestrate other automated tasks using templates to roll out all the required components ofExchange 2010 as it scales.

This building block approach reduces the time to deploy Exchange 2010 to production, and ensureshardware and software policies are applied, even if Exchange 2010 is moved to new hardware. As

business requirements change, IT departments can adjust their resource allocation strategyseamlessly and predictably by using the Vblock platforms scaling model to provision, or even

reallocate, compute and storage resources.

Scope

The solution architecture describes the infrastructure for Exchange 2010 with the following

requirements:

20,000 mailboxes

2 GB mailbox size

150 messages sent and received per mailbox per day

100% user concurrency

100% user capacity

2 copies of each database within a single Database Availability Group (DAG), defined in Table 1, and

within a single data center

The JetStress and Load Generator tests measures performance and reliability for 20,000 mailboxesusing four mailbox servers, with 5,000 mailboxes per server. Enterprises have significant storage


4/27


requirements; therefore, testing was performed to show how Vblock platforms can easily scale inincrements of 5,000 to 20,000 mailboxes.

ObjectivesThe objectives for this Solution Architecture are to:

Design and operate an entire Exchange 2010 infrastructure on a virtualized Vblock platform

Demonstrate the performance goals and results in context with the requirements, starting with 5K

mailboxes and scaling to 20K mailboxes, on the Vblock Series 700 model MX

Provide the Exchange 2010 sizing process on the Vblock 700 with the requirements outlined

Validate the end to end performance requirements at 100% capacity and 100% concurrency, using

Microsoft JetStress 2010 and Load Generator 2010

Adhere to the best practices for designing Exchange 2010 per VMware, EMC, and Microsoft

recommendations, such as the separation of Exchange Server 2010 High Availability (HA), defined in

Table 1, from the VMware HA and DRS functionality

Leverage a virtual provisioning strategy for storage design

AudienceThis document is intended for technical engineering staff, managers, IT planners, administrators, and

other IT professionals involved in evaluating, managing, operating, or designing Exchange 2010deployments on the Vblock platforms.

Terminology

Table 1: Terminology for the solution architecture

Term Definition

BackgroundDatabase

Maintenance(BDM)

Process of the Exchange 2010 database maintenance that involvesperforming a checksum value on both active and passive database copies.

High Availability(HA)

The concept of providing data and service resiliency in the event of anoutage within a single data center, regarding a specific solution or

infrastructure

Database

Availability Group(DAG)

A DAG is an Exchange 2010 cluster, and is intended to provide for high

availability and site resiliency for Exchange 2010 mailbox servers. A DAGcan contain up to 16 mailbox servers. The servers host databases and can

provide automatic database-level recovery from failures affecting individualdatabases. Any server in a DAG can host a copy of a mailbox database

Regular devices Represent fully-provisioned, traditional LUNs (also known as thick

devices). In this conventional storage provisioning model, space isallocated in excess of current storage requirements in anticipation ofexpected growth.

Virtual Provisioning An industry standard technology whereby a large virtual device (that is, avolume) is configured and presented to the host while consuming physical

storage from a shared pool only as needed.EMC's Virtual Provisioning technology relies on the on-demand allocation

of blocks of data versus the traditional method of allocating all the blocks upfront. This methodology eliminates whitespace, which helps avoid the poorutilization rates that occur in the traditional storage allocation method (full

provisioning) where large pools of storage capacity are allocated to


5/27


individual servers but remain unused (not written to).

Virtual Machine

(VM)

A virtualized instance of an operating system that a virtualization host and

hypervisor manage and abstract the virtual machine from the underlyingphysical hardware resources

Solution Architecture

Vblock Platform OverviewThe Virtual Computing Environment Company (VCE) has created an integrated, validated Vblock

infrastructure platform, built from best-of-class components for compute, network, storage, andvirtualization.

Vblock Infrastructure Platforms are enterprise and service provider-class IT infrastructure that are pre-engineered, tested, and validated units with a defined performance, capacity, and availability Service

Level Agreement (SLA). Vblock Infrastructure Platforms deliver IT infrastructure in a new way andaccelerates organizations migration to the cloud. They simplify IT infrastructure acquisition,

deployment, and ongoing operations.

The Vblock 700 is enabled by Cisco Unified Computing system, EMC Symmetrix VMAX, and VMwarevSphere 4.1, and delivers high-performance and large-scale computing infrastructures.

Vblock platforms are characterized by:

Repeatable units of infrastructure based on performance, operational characteristics, and requirements of

power, space, and cooling

Repeatable design patterns that facilitate rapid deployment, integration, and scalability

Architecture that can be scaled for the highest efficiencies in virtualization of workloads

An extensible management and orchestration model based on industry-standard tools, APIs, and methods

An architecture that contains, manages, and mitigates failure scenarios in hardware and software

solutions


6/27


Solution Components

The following table lists the Vblock 700 components as tested.

Table 2: Vblock 700 category components

Category Components

Compute Cisco 6120 or 6140 Fabric InterconnectsB250-M2 Blades

UCS can boot the OS from the SAN or from local storage

Network The UCS uses 6100 series fabric interconnects that carry the networkand storage traffic from the blades to the connected SAN and LAN

Cisco Nexus 5000 data center switches for access layer, and Cisco6500 series data center switches for aggregation layer.

Storage EMC Symmetrix VMAX SAN storageCisco MDS SAN director

Virtualization VMware vSphere/vCenter 4.1

Management EMC Ionix Unified Infrastructure Manager (UIM) 1.0VMware vCenter 4.1

EMC Symmetrix Management consoleEMC PowerPath/VE

Cisco UCS ManagerCisco Fabric Manager

Software Microsoft Exchange Server 2010 SP1

Windows Server 2008 R2Active Directory

The Vblock 700 provides template-based virtual provisioning. Using template-based virtual

provisioning an enterprise can:

Enable the rapid deployment of resources, reducing operational expenses

Reduce configuration errors through automation of resource allocation

Provision storage on an as needed basis

Allow an organization to scale at different levels by leveraging various templates that contain pre-

configured resources geared for the varied scaling scenarios.


7/27


The solution components listed above, and discussed throughout the following sections, are displayedin Figure 1, below.

Figure 1. Logical Design: Exchange 2010 on Vblock 700

The following sections identify and describe the Vblock 700 components used in the solutionarchitecture.

Cisco Unified Computing System (UCS)The Cisco Unified Computing System is a next-generation data center platform that unites compute,

network, storage access, and virtualization into a cohesive system designed to reduce Total Cost of

Ownership and to increase business agility. The system integrates a low-latency, lossless 10 Gigabit

Ethernet unified network fabric with enterprise-class, x64-architecture servers. The system is anintegrated, scalable, multi-chassis platform in which all resources participate in a unified managementdomain.

Table 3, below, describes the main system components of the Cisco Unified Computing System

(UCS). Working as a single, cohesive system, these components unify technology in the data center.They represent a radical simplification in comparison to traditional systems, helping to simplify datacenter operations while also reducing power and cooling requirements. The system amplifies IT agility

for improved business outcomes.


8/27


Table 3: Cisco Unified Computing System components

Component Description

Compute Blade servers based on Intel Xeon 5500, 5600, and 7500 seriesprocessors. The blade servers offer patented Cisco Extended Memory

Technology to support applications with large datasets and allow morevirtual machine density per blade

Network Low-latency, lossless, 10 Gbps unified network fabric that consolidatesEthernet and SAN fabric networks into one. The unified fabric reduces

the number of network adapters, switches, and cables, and decreasesthe power and cooling requirements.

Virtualization Virtualization unleashes the full potential of hardware by maximizing

the scalability, performance, and operational control of virtualenvironments. Cisco security, policy enforcement, and diagnostic

features are now extended into virtualized environments to better

support ever changing business and IT requirements.Storage access Provides consolidated access to both SAN storage and network

attached storage (NAS) over the unified fabric. Unifying storage accessmeans that the Cisco Unified Computing System can access storageover Ethernet, FC, FC over Ethernet (FCoE), and iSCSI, providing

customers with choice and investment protection. Administrators canpre-assign storage-access policies for system connectivity to storage

resources, simplifying storage connectivity and management whilehelping to increase productivity.

Management Integrates all the system components to manage the entire solution as

a single entity through the Cisco UCS Manager software. Cisco UCSManager provides an intuitive graphical user interface (GUI), a

command-line interface (CLI), and a robust application-programming

interface (API) to manage all system configuration and operations.Cisco UCS Manager helps increase IT staff productivity, enabling

storage, network, and server administrators to collaborate on definingservice profiles for applications. Service profiles are logical

representations of desired physical configurations and infrastructurepolicies. They help automate provisioning and increase agility, allowing

for the provisioning of resources in minutes instead of days.

Each UCS server used is a full width blade with a mezzanine-connected converged network adapter(CNA) for up to 20 Gbps of I/O throughput. The CNA provides dual-port connectivity to the server

chassis midplane and presents the host operating system with two 10 Gigabit Ethernet adapters andtwo QLogic 4-Gbps FC HBAs.

Each UCS chassis includes two fabric extenders. A fabric extender has four 10-Gigabit Ethernet ports,which are, FCoE-capable, with several Small Form-Factor Pluggable Plus (SFP+) ports that connect

the blade chassis to the Fabric Interconnect, and eight 10-Gigabit ports connected through the mid-plane to each half-width slot in the chassis. The two fabric extenders provide redundancy and up to 80

Gbps of I/O to the chassis.


9/27


Table 4 provides information about the Cisco hardware used in this solution.

Table 4: Cisco Unified Compute SystemItem Description

Blade Server B250 M2

Processors 2 x Intel Xeon x5680 (3.33GHz)

Memory 256 GB RAM per blade

Converged Network Adapter M81KR (VIC) Adapter

Blade Storage Boot from SAN

Chassis 5108 (6RU)

Fabric Interconnect 2 x 6120XP

Fabric Interconnect Expansion Module 2 x 8-port 4 Gb/s FC

EMC Symmetrix VMAXThe Symmetrix VMAX is built on the strategy of simple, intelligent, modular storage and incorporatesa new Virtual Matrix interface that connects and shares resources. This allows the storage array toseamlessly grow from an entry-level configuration into the worlds largest storage array that provides

the highest levels of performance and availability.

The following summarizes the VMAX configuration used in this solution:

Table 5: VMAX configuration

Item Description

Engines 4

SATA hard drives using RAID 1/0 238 1 TB 7.2k rpm

Global Memory 256 GB (64 GB on each engine)

FA Ports 8 x 4GB

EMC Symmetrix Virtual ProvisioningAn integrated Symmetrix array feature enabling organizations to present a certain amount of virtual

capacity to a host, allowing it to consume storage as needed.

Virtual provisioning:

Can improve storage capacity utilization by reducing the need to provision new storage frequently, and

helps avoid the cost associated with allocated but unused storage

Simplifies the data layout with automated wide striping, enabling organizations to achieve equivalent or

better performance than standard storage provisioning, with less planning and effort required

Automates storage pool rebalancing, which allows users to nondestructively balance workloads andextend storage pool capacity in small increments if required, while maximizing performance

Ultimately, this technology allows an organization to reduce their total cost of acquisition andownership of storage.

Virtual Provisioning was used in the VMAX configuration, as it is a recommendation from EMC forExchange 2010 solutions.


10/27


For the Virtual Provisioning storage pool:

238 1TB SATA disks in total are used across 4 x Engines

476 x 228 G RAID 1 devices (TDATs) in total are created as the data device and put into one storage

pool. This provides a 100 TB storage pool.

For the Exchange 2010 environment (TDEVs):

64 x 900G thin devices in total (each device is a 10-way meta device containing 10 x 90G members) are

created for Exchange 2010 servers used as DB LUNs. 64 x 90G thin devices in total are created for

Exchange 2010 as Log LUNs. This is provisioned so that each Exchange 2010 DB server has 16 x 900GDB LUNs, and 16 x 90G log LUNs, for a total of 4 x Exchange 2010 DB servers.

2 x 900G and 2 x 90G thin devices in total are created and used for two Exchange 2010 HUB/CAS

servers, each has 1 x 900G thin LUN and 1 x 90G thin LUN

The above thin LUNs are bound to the storage pool created above.

EMC has done extensive work with virtual provisioning in the VMAX platform, and more detail canbe obtained from their white paper that is listed in the reference section of this document.

Exchange 2010 SP1This section covers the Microsoft components of the solution: Exchange 2010 SP1 Enterprise Edition,

Windows Server 2008 R2, and Active Directory. The Exchange 2010 server roles used in this solution

are the mailbox role, hub transport role, and the client access server (CAS) role. All Exchange 2010roles are virtualized using VMware ESXi 4.1.

Microsoft has created a program called Server Virtualization Validation Program (SVVP) for the

purpose of providing full support for Windows Operating Systems running in a virtualized environment.The goal and outcome of the certification program is to ensure that Microsoft can provide full supportfor the OS and Microsoft applications sitting on top of the OS when virtual guests are hosted on

different hypervisors. This solution is running on Windows Server 2008 R2 with VMware ESXi 4.1.This combination is certified by the SVVP program, and customers can expect Microsoft to provide full

supportability of the Windows 2008 R2 and Exchange 2010 SP1 applications in this solution.

The different Exchange 2010 roles and their respective subsystems were sized as follows:

Mailbox Role

Processor

o The megacycles required per mailbox are 3

o The megacycles required for a 4 core mailbox server is 16500

o The specInt2006 ratio for the Cisco B250 M2 blades is estimated at 380, based on the Intel Xeon 5600

processor architecture

o 10% processor overhead was factored into the overall requirements to compensate for the hypervisor

overhead, per VMware guidance.

o With this configuration, the server would run at 74percent utilization

o The Microsoft mailbox role calculator was used to determine the processor requirements

Storage

o The EMC VMAX calculator recommends 212 1GB 7.2K SATA drives to meet the IO requirements of

20,000 users with a 150 messages sent/received per mailbox per day, with a 2GB mailbox quota.


11/27


Single item retention and calendar version storage is included in this calculation. The IOPS per mailbox

is .18. There is no IOPS or disk capacity buffer included.

o The Microsoft Exchange 2010 Mailbox Role calculator recommends 240 1GB 7.2K SATA drives to meet

the IO requirements of 20,000 users with a 150 messages sent/received per mailbox per day, with a

2GB mailbox quota. Single item retention and calendar version storage is included in this calculation.

The IOPS per mailbox is .15. There is no IOPS or disk capacity buffer included.

o The Vblock 700 VMAX virtual storage pool was configured with 238 1GB 7.2K SATA, and successfully

passed the JetStress test goals of 900 IOPS.

o RAID 1/0 was the best fit for capacity and IO with this configuration.

Memory

o The amount of DB cache required per mailbox is 9 MB. The amount of memory required per server is64 GB

o The Microsoft mailbox role calculator was used to determine the memory requirements

Summary of each mailbox server

o 4 vCPUs

o 64 GB RAM

o 4 x vHBAs

o 32 LUNs

o 16 900 GB DB LUNs

o 16 90 GB LOG LUNs

o 2 x vNICs

Notes

o The backup and restore of Exchange 2010, outside of a JetStress test, is not covered in this solution

architecture. EMC has a proven and documented backup and restore solution using Replication

Manager. A document link is provided in the References section.

Hub Transport and Client Access Servers

Processor

o A 1:1 ratio was used for mailbox cores to combined hub transport and client access server cores. 16

mailbox cores means 16 hub transport and client access server cores.

Storage

o The client access server role does not consume much storage space, and does not require a certain

amount of IOPS or capacity. The Hub transport role does have this requirement, based on the message

volume per day to determine the IOPS and capacity required. There are other factors that influence the

Hub transport storage capacity requirements, such as the transport dumpster and shadow redundancy. Memory

o Memory was sized by using 2GB of memory for every processor in the system, following Microsoft

recommendations for memory starting points for hub transport and client access server combinations.

Summary of each HUB/CAS server

o 6 vCPUs

o 12 GB RAM

o 2 x vNICs


12/27


Notes

o Hardware or software load balancing was not used at the CAS tier in the solution testing. Although a

real Exchange 2010 solution would include this, it is outside the scope of this solution testing.

Domain Controllers/Global Catalogs

Processor

o 3 processor cores are recommended to support the 20,000 mailboxes per the Microsoft role calculator.

4 cores were used for each virtual machine, and there were 2 DCs/GCs used in this solution

Storage

o100GB LUNs were used for storage

Memory

o 4 GB of RAM was used

Summary

o 4 vCPUs

o 4 GB RAM

o 2 x vNICs

Notes

o A single Active Directory site is used, with two Domain Controllers and Global Catalogs.


13/27


The following diagram illustrates the virtual storage configuration on the VMAX.

Figure 2. Disk Layout

VMware vSphere

VMWare ESXi 4.1, update 1, was used for three virtualization hosts. VMWare vCenter server 4.1,update 1, was used for managing the virtual infrastructure in this solution. EMC PowerPath/VE was

used for multipathing and load balancing.


14/27


Solution Validation

OverviewThis section describes the solution validation and test results. Exchange 2010 solutions are validated

using Microsoft JetStress 2010 (JetStress) and Microsoft Load Generator 2010 (Load Generator).

JetStress provides four tests that stress different IO patterns experienced by Exchange 2010. These

tests are covered in greater detail in Appendix A. JetStress takes input and based on therequirements creates databases and tests these databases from a extensible storage engineperspective (the database engine on which Exchange 2010 runs).

After successfully completing the JetStress disk performance and stress tests in a non-productionenvironment, you will have ensured that your Exchange 2010 disk subsystem is adequately sized forperformance and capacity requirements.

Load Generator tests the client facing components of the Exchange 2010 solution, and tests theprocessor and memory subsystems of the Exchange 2010 infrastructure.

Load Generator is a simulation tool to measure the impact of MAPI, OWA, IMAP, POP and SMTPclients on Exchange 2010 servers. Load Generator is used to validate the end to end e-mailinfrastructure.

Together, the JetStress and Load Generator tools are used as an industry standard to validate

Exchange 2010 solutions and the hardware on which they are run.

The following sections summarize both the JetStress and Load Generator test results. More detailabout both tests can be found in the appendices A and B.


15/27


Summary

This section contains a summary of the test results. The successful JetStress results and LoadGenresults validate at 100% capacity, and 100% concurrency, the end to end performance requirementsof the Exchange 2010 and Vblock 700 solution.

See Appendices A and B for test details.

Table 6: Summary of the four JetStress test results

JetStress2hr Test

JetStress24hr Test

DatabaseBackup Test

Soft RecoveryTest

Database I/Os

Total disk transfers/sec 924.512 924.848 - -

Total disk reads/sec 547.088 545.92 - -

Total disk writes/Sec 377.424 378.928 - -

Avg. DB disk read latency 11.838 ms 12.127 ms - -

Avg. DB disk write latency 9.226 11.321 - -

Transactional log I/Os

Total log disk writes/sec 320.944 325.296 - 28.173

Avg. log disk write latency 0.812 0.818 - 0.443 ms

Total BDM IO reads/sec 477.504 494.464 - -

Total IOPS Per MailboxVM 1722 IOPS 1744.608

MB reads/sec/database - - 59.8 -

MB reads/sec per server - - 956.47 -

Avg. time to play one logfile - - - 5.3 Sec

Table 7: Summary of the Load Generator test results

Parameter Target Tested results

Messages Delivery Rate / Mailbox 0.34 0.34

IOPs / mailbox 0.15 0.18

Megacycles / Mailbox 3 3.35

Further details are presented in Appendices A and B.


16/27


Conclusion

The Solution architecture validation demonstrates that the following key results have been realized

and the objectives of this solution architecture have been met.

Design and operate an entire Exchange 2010 infrastructure on a virtualized Vblockplatform:The Solution Architectureand Componentssections describe the infrastructure design on the Vblock

Platforms. The test results in Appendices A and B demonstrate the successful operation of the

Exchange 2010 infrastructure on the virtualized platform through:

Disk performance tests

Stress tests to prove that the sizing is adequate for performance and capacity requirements

Tests of the client facing components, processor and memory subsystems Tests of the client impacts on the Exchange 2010 servers

Tests to validate the entire e-mail infrastructure

The virtualized Exchange Server 2010 design approach described in this paper is easy to use andcan scale as needed to meet an organization's needs and requirements.

Using a Vblock 700 configuration, larger numbers of Exchange 2010 IOPS can be produced incomparison to Direct Attached Storage configurations, resulting in fewer spindles needed to produce

the desired performance.

Vblock platforms provide impressive management capabilities. EMC Ionix Unified Infrastructure

Manager optimizes the Vblock 700 infrastructure by streamling, standardizing, and automating theconfiguration and provisioning of the Vblock platform. This helped reduce operational costs, improveresponsiveness to the needs of the business, and scaled the infrastructure without increasing

headcount.

Other benefits of running Exchange 2010 on the Vblock Platforms include:

The capability to rapidly provision your Exchange 2010 infrastructure using Unified Infrastructure Manager

(UIM)

The ability to leverage the VMware features, such as:

o vCenter Chargeback, which maps infrastructure costs to business units, cost centers, or external

customers.

o CapacityIQ, which helps determine resource utilization and help with capacity monitoringo VMware HA, VMware vMotion, and DRS for the HUB and CAS roles, as well as other servers that are

not clustered using Microsoft Failover Clustering.

Demonstrate the performance goals and results in context with the requirements,

starting with 5K mailboxes and scaling to 20K mailboxes, on the Vblock 700:Based on successful JetStress results provided in Appendix A, and successful LoadGen resultsdocumented in Appendix B, the data summarizations of both sets of tests are featured in the

Summary section. These successful test results helped validate the 20,000 user building blocks byspreading loads in blocks of 5,000 users across four virtualized mailbox servers running on a singleblade. It helped achieve a 4:1 consolidation ratio.


17/27


Implementation of Virtually Provisioned Storage helped with on demand Exchange 2010 databasegrowth, including deployments of additional virtualized Exchange 2010 mailbox servers without

causing any type of outages.

Provide the Exchange 2010 sizing process on the Vblock 700:The Exchange 2010 sizing process was defined based on the results from Microsofts Role Calc and

EMCs Exchange Storage Calc. Microsoft and EMCs best practice guidelines were adhered to during

our design and testing of this Exchange 2010 Solution Architecture.The Exchange 2010 sizingprocess is shared throughout the paper: Microsoft Sizing guidelines are followed for processor,

memory, and storage sizing, Figure 2 describes the storage disk layout arrangement, and examplesare provided in the References section with the link to a sizing paper Exchange on VMware Sizing

Examples.

Validate the end-to-end performance requirements at 100% capacity and 100%

concurrency, using Microsoft JetStress 2010 and Load Generator 2010:

The successful JetStress results provided in Appendix A, and successful LoadGen resultsdocumented in Appendix B, validate at 100% capacity and 100% concurrency the end to endperformance requirements.

Adhere to the best practices for designing Exchange 2010 per VMware, EMC, andMicrosoft recommendations, such as the separation of Exchange Server 2010 HA

from the VMware HA and DRS functionality:Appendix C provides a detailed description of the best practices to follow for running Exchange 2010

on the Vblock Platforms.

Leverage a virtual provisioning strategy for storage design:The EMC Symmetrix Virtual Provisioningsection provides virtual storage provisioning details,

discusses the benefits of the strategy, and describes how the virtual provisioning was done on theVMAX.


18/27


References

VMware

Exchange 2010 on VMware Best Practices Guide

Exchange 2010 on VMware Design and Sizing Examples

EMC

EMC Virtual Infrastructure for Microsoft Exchange 2010 Enabled by EMC Symmetrix VMAX, VMware

vSphere 4 and EMC Replication Manager

Performance, and Availability at Scale Enabled by EMC VMX, VP and VMware vSphere

Microsoft

Understanding Processor configurations and Exchange Performance

Understanding Memory configurations and Exchange Performance

Mailbox Storage Design Process

Mailbox Server Processor Capacity Planning

Exchange 2010 Mailbox Server Role Requirements Calculator

Performance and Scalability Counters and Thresholds
http://www.vmware.com/files/pdf/exchange-2010-on-vmware-best-practices-guide.pdfhttp://www.vmware.com/files/pdf/exchange-2010-on-vmware-best-practices-guide.pdfhttp://www.vmware.com/files/pdf/exchange-2010-on-vmware-best-practices-guide.pdfhttp://www.vmware.com/files/pdf/exchange-2010-on-vmware-best-practices-guide.pdfhttp://www.vmware.com/files/pdf/exchange-2010-on-vmware-design-and-sizing-examples.pdfhttp://www.vmware.com/files/pdf/exchange-2010-on-vmware-design-and-sizing-examples.pdfhttp://www.vmware.com/files/pdf/exchange-2010-on-vmware-design-and-sizing-examples.pdfhttp://www.vmware.com/files/pdf/exchange-2010-on-vmware-design-and-sizing-examples.pdfhttp://powerlink.emc.com/km/live1/en_US/Offering_Basics/White_Paper/h7392-virtual-exchange-vmax-vsphere-psg.pdf?mtcs=ZXZlbnRUeXBlPUttQ2xpY2tDb250ZW50RXZlbnQsZG9jdW1lbnRJZD0wOTAxNDA2NjgwNTUzZWFmLG5hdmVOb2RlPTBiMDE0MDY2ODAxZWE1Yjg_http://powerlink.emc.com/km/live1/en_US/Offering_Basics/White_Paper/h7392-virtual-exchange-vmax-vsphere-psg.pdf?mtcs=ZXZlbnRUeXBlPUttQ2xpY2tDb250ZW50RXZlbnQsZG9jdW1lbnRJZD0wOTAxNDA2NjgwNTUzZWFmLG5hdmVOb2RlPTBiMDE0MDY2ODAxZWE1Yjg_http://powerlink.emc.com/km/live1/en_US/Offering_Technical/White_Paper/h7018-exchange-vmax-vp-vsphere-wp.pdf?mtcs=ZXZlbnRUeXBlPUttQ2xpY2tDb250ZW50RXZlbnQsZG9jdW1lbnRJZD0wOTAxNDA2NjgwNGY5NWM3LG5hdmVOb2RlPTBiMDE0MDY2ODAxZWE1Yjg_http://powerlink.emc.com/km/live1/en_US/Offering_Technical/White_Paper/h7018-exchange-vmax-vp-vsphere-wp.pdf?mtcs=ZXZlbnRUeXBlPUttQ2xpY2tDb250ZW50RXZlbnQsZG9jdW1lbnRJZD0wOTAxNDA2NjgwNGY5NWM3LG5hdmVOb2RlPTBiMDE0MDY2ODAxZWE1Yjg_http://technet.microsoft.com/en-us/library/dd346699.aspxhttp://technet.microsoft.com/en-us/library/dd346699.aspxhttp://technet.microsoft.com/en-us/library/dd346700.aspxhttp://technet.microsoft.com/en-us/library/dd346700.aspxhttp://technet.microsoft.com/en-us/library/ff367907.aspxhttp://technet.microsoft.com/en-us/library/ff367907.aspxhttp://technet.microsoft.com/en-us/library/ee712771.aspxhttp://technet.microsoft.com/en-us/library/ee712771.aspxhttp://msexchangeteam.com/archive/2009/11/09/453117.aspxhttp://msexchangeteam.com/archive/2009/11/09/453117.aspxhttp://technet.microsoft.com/en-us/library/dd335215.aspxhttp://technet.microsoft.com/en-us/library/dd335215.aspxhttp://msexchangeteam.com/archive/2009/11/09/453117.aspxhttp://technet.microsoft.com/en-us/library/ee712771.aspxhttp://technet.microsoft.com/en-us/library/ff367907.aspxhttp://technet.microsoft.com/en-us/library/dd346700.aspxhttp://technet.microsoft.com/en-us/library/dd346699.aspxhttp://powerlink.emc.com/km/live1/en_US/Offering_Technical/White_Paper/h7018-exchange-vmax-vp-vsphere-wp.pdf?mtcs=ZXZlbnRUeXBlPUttQ2xpY2tDb250ZW50RXZlbnQsZG9jdW1lbnRJZD0wOTAxNDA2NjgwNGY5NWM3LG5hdmVOb2RlPTBiMDE0MDY2ODAxZWE1Yjg_http://powerlink.emc.com/km/live1/en_US/Offering_Basics/White_Paper/h7392-virtual-exchange-vmax-vsphere-psg.pdf?mtcs=ZXZlbnRUeXBlPUttQ2xpY2tDb250ZW50RXZlbnQsZG9jdW1lbnRJZD0wOTAxNDA2NjgwNTUzZWFmLG5hdmVOb2RlPTBiMDE0MDY2ODAxZWE1Yjg_http://powerlink.emc.com/km/live1/en_US/Offering_Basics/White_Paper/h7392-virtual-exchange-vmax-vsphere-psg.pdf?mtcs=ZXZlbnRUeXBlPUttQ2xpY2tDb250ZW50RXZlbnQsZG9jdW1lbnRJZD0wOTAxNDA2NjgwNTUzZWFmLG5hdmVOb2RlPTBiMDE0MDY2ODAxZWE1Yjg_http://www.vmware.com/files/pdf/exchange-2010-on-vmware-design-and-sizing-examples.pdfhttp://www.vmware.com/files/pdf/exchange-2010-on-vmware-best-practices-guide.pdf


19/27


Appendix A JetStress 2010 Results

Four JetStress tests were run, as described in the following table.

The test results described in this appendix validate the storage, demonstrating that the storage meetsthe performance goals.

Table 1A: JetStress Tests

Test Test name Description

1 JetStress 2-hour performance test Measures baseline storage I/O performanceand how long it takes for the storage to respondto an I/O under load.

2 JetStress 24-hour stress test Validates how storage responds to a high I/Oload for an extended period of time.

3 JetStress database backup VSS test Measures the maximum rate at whichdatabases can be backed up via VSS.

4 JetStress soft recovery test Measures the maximum rate at which the logfiles can be played against the databases.


20/27


The variables used for the tests include:

Table 2A: JetStress test configuration

Item Value

Exchange 2010 Mailboxes Simulated 20,000 Mailboxes

ESXi hosts 1 Physical Server

Virtual Machines per ESX Server 4

DB copies 2

Active Mailboxes per Server 5000

Databases per host 64

Mailboxes per database 312

Mailbox size 2GB

User profile150 messages sent and received per dayper mailbox

Database Read/Write Ratio 3:2

Database Maintenance Configuration 24x7 BDM

Database LUN size 900 GB

Log LUN Size 90 GB


21/27


JetStress two-hour storage test performance results

Primary storage performance testing is designed to exercise storage with the maximum Exchange

2010 type of I/O for a period of two hours. The test is designed to show how long it takes for thestorage to respond to an I/O under load.

(1) Mailbox VM Performance Results (5,000 users)

Database I/Os Avg. performance per Mailbox VM (5,000 users)

DB disk transfers/s 924.512

DB disk Reads/s 547.088

DB disk Writes/s 377.424

Avg DB disk Read latency 11.838

Avg DB disk Write latency 9.226


Log disk Writes/s 320.944

Avg log disk Write latency 0.812ms

BDM IOPS

Total BDM IOPS 477.504

Total IOPS per Mailbox VM 1722 IOPS

Two-hour (16:00-18:00) totalI/O throughput at 15 min intervals:

Figure 3. 2-hour (16:00-18:00) total I/O throughput at 15 minute intervals


22/27


Soft Recovery test: Transaction log recovery/replay performance

The Soft Recovery JetStress test measures the maximum rate at which the log files can be played

against the databases. The following table shows test results for the average rate for 500 log filesplayed in a single database. Each log file is 1 MB in size.

(2) Soft Recovery PerformanceAverage time to play one log file 5.3 sec

Database backup and recovery performance overview

Two JetStress tests were performed with a single ESX server and 20,000 users to measure storagebackup performance, as follows:

Database Read-only Performance measures the sequential read rate of the database files. Transaction Log Recovery/Replay Performance measures the recovery/replay performance by playing

transaction logs into the database.

The backup VSS JetStress test measures the maximum rate at which databases can be backed upusing VSS. The following table shows test results for the average rate for a single database file.

(3) Average Rate for Single Database File

MB read/s per database 59.8

MB read/s total per server 956.47

Note

DB recovery and Read-only performance tests results are the same for both configurations and arecombined in a single table for easy review

(4) Mailbox VMs Performance Results (20,000 users)

Database I/Os Avg performance per ESX server(4 Mailbox VMs/20,000 users)

DB disk transfers/s 3699

DB disk reads/s 2188.352

DB disk writes/s 1509.696

Avg DB disk read latency 11.838ms

Avg DB disk write latency 9.226ms


Log disk writes/s 1283.776Avg log disk write latency 0.812ms

BDM IOPS

Total BDM IOPS 1910 (261 K)

Total Transactional IOPS 6892.776


23/27


Appendix B Load Generator 2010 Results

After completing the storage validation with JetStress and determining that the storage meets theperformance goals, the next step in the validation process is to use Load Generators 2010 (LoadGen)to simulate client workload against the end to end Exchange 2010 infrastructure. LoadGen testing isnecessary to determine how each Exchange 2001component performs under a simulated product

load.

LoadGen requires a complete Exchange 2010 infrastructure. You should perform all LoadGenvalidation testing using a non-production environment LoadGen creates users, groups, OUs,

Exchange 2010 databases, and mailboxes, and executes workload testing against the Exchange2010 HUB, CAS, and Mailbox roles, as well as the network and storage components.

In this test, LoadGen is used to simulate mailboxes using Outlook 2007 online mode with the followingcharacteristics:

The action profile is 150 messages sent/received per mailbox per day

Each mailbox is 2 GB in size

Each database contains 313 mailboxes

To simulate a normal operation, the simulated workload duration is set to eight hours and each

simulation runs for eight hours.

The 150-messages profile sends 50 messages and receives 100 messages per mailbox per day. It isexpected that during an eight hour simulated day the Mailbox server with 5000 active users will log

approximately 8.7 sent messages per second and 17.4 received message per second. Use theformula below to calculate the expected number of sent and received messages per second.

Messages sent per second

Messages received second

Mailbox server response times for client requests are tracked to determine the amount of time ittakes a mailbox server to respond to a client request. The response time average per requestshould not exceed 10 milliseconds on average. Use the following system monitor counters on themailbox role to monitor response time.

MSExchangeIS\RPC Averaged Latency

MSExchangeIS Mailbox\RPC Averaged LatencyMSExchangeIS Client (*)\RPC Average Latency

As a best practice, it is recommended to disable Hyper Threading on the ESX hosts. This requires

entering the BIOS, disabling hyper threading, and doing a cold shutdown of the server. The validity ofthe each test is determined by comparing the results of the performance counters to Microsoft

performance thresholds. Performance counter data is collected at 10 second intervals for the durationof each test run. The results of the first and last hours are discarded. Results are averaged over theremaining duration of the test.


24/27


Table 1B lists the primary counters and their validation criteria:

Table 1B: Primary counters and validation criteria

Performance monitor counter Criteria

Processor(Total)\% Processor Time Not to exceed 80% during peak load

MSExchangeIS\RPC Averaged Latency Not to exceed 10 msec

MSExchangeIS Mailbox(Total)\Messages Sent\sec Approximately 0.002083messages/second/mailbox

MSExchangeIS Mailbox(Total)\Messages Received\sec Approximately 0.008333messages/second/mailbox

Logical Disk Sec/read Not to exceed 20 msec

Logical Disk Sec/write Not to exceed 20 msec

For additional information about monitoring Exchange 2010 performance and other key performance

counters, visit Performance and Scalability Counters and Thresholds on Microsofts TechNet site. A

link is provided in the References section.

Table 2B lists the LoadGen test that the team ran to measure the performance of the Exchange 2010

Infrastructure.

Table 2B: Environment validation tests with LoadGen

Test Description

1 During heavy operations at peak load, this test included a 100% concurrency test with 150Outlook messages online with a MAPI profile.

Table 3B lists the test of heavy operating conditions at peak load as compared to the normal operating

conditions.

Table 3B: Test of Heavy operation condition with peak load

Test Description

Test Objective In this test, the objective was to validate the entire Exchange 2010 environmentunder heavy operating condition with the peak load. ESXi host and VMs performancewere measured against Microsoft recommended performance targets.

Test configuration In this test, all Exchange 2010 VMs were under normal operating condition. LoadGenwas configured to simulate heavy user profile (150), which doubles the send andreceive messages per second.

Test Performance results and analysis:

Table 4B shows results achieved with Exchange 2010 VMs within the target goals. During peak loads,

the average CPU utilization on the Mailbox VMs was approximately 65 percent and the UCS ESXi

host utilization was about 68 percent. On the CAS/HUB VMs, CPU utilization was approximately 38percent.


25/27


Table 4B. Validation of expected load for test

Parameter Target Tested results

Messages Delivery Rate / Mailbox 0.34 0.34

IOPs / mailbox 0.15 0.18

Megacycles / Mailbox 3 3.35


26/27


Appendix C Best Practices

This appendix covers some of the best practices to follow for running Exchange 2010 on the VblockPlatforms. This is not meant to be an exhaustive list of best practices, but some of the key ones toconsider.

UCS best practices for Exchange 2010:

Implement redundant components to avoid downtime in case of component failure. Optional redundancy

is available for the following components:

o Fabric interconnect

o Fabric interconnect power supply

o Fabric extender

o Chassis power supply

Configure dual HBA adapter ports and PowerPath for storage path redundancy. Configure one HBA portfor Fabric A, and another HBA for Fabric B.

Configure the Fabric Interconnects for end-host-mode. This option enables fabric failover for LAN

adapters.

Configure dual network adapters for LAN access redundancy. Configure one adapter to use fabric A and

another adapter to use fabric B. Enable both adapters for fabric failover. You do not need to deploy

network adapter teaming software in this case.

Configure redundant network links for the fabric interconnect uplink Ethernet ports using virtual port

channel (vPC) or similar technologies.

Separate DAG replication traffic from client MAPI traffic on different adapters. Configure one adapter for

database replication and seeding, and a different adapter for MAPI traffic. Each adapter resides on its own

subnet. Use UCS Manager service profile pools to allocate WWPN, WWNN, MAC addresses, and server

UUIDs. This option enables these attributes to move from one server blade to another server blade with

the service profile. One benefit of this approach is a case where a blade server fails and needs to be

replaced with new hardware. You can replace the blade server without reconfiguring the SAN zoning, LUN

masking, VLAN assignments, or static DHCP reservations.

VMware Best practices for Exchange 2010:

ESXi hosts were clustered; however vMotion, DRS and HA were not used for the Exchange 2010 mailbox

role. VMware does not recommend enabling these features for servers that are part of a Failover Cluster,

hence there is no mixing of VMware HA and Exchange HA.

Do not oversubscribe virtual processor cores to physical processor cores more than 2:1 (vCPU:pCPU).

Processor oversubscription is an estimate around the concurrency of Exchange 2010 load. VMware

recommends a 1:1 ratio for high workloads or where the concurrency fluctuates greatly or is not known forcertain.

EMC Best practices for Exchange 2010:

If running Exchange 2010 is exclusively running on the VMAX platform, then do not use Flash disk.

Microsoft best practices for Exchange 2010:

Follow the sizing guidelines for storage, processor and memory, per the current TechNet articles for

Exchange 2010 SP1 (links are provided in the references section of this document).


27/27

Design your systems as if they were running on bare metal hardware and then port yourrequirements over to virtualization following virtualization guidelines accordingly.

Storage calculation: always calculate spindles for performance first, and then for capacity.

ABOUT VCEVCE, the Virtual Computing Environment Company formed by Cisco and EMC with investments from VMware and Intel,

accelerates the adoption of converged infrastructure and cloud-based computing models that dramatically reduce the

cost of IT while improving time to market for our customers. VCE, through the Vblock platform, delivers the industry's first

completely integrated IT offering with end-to-end vendor accountability. VCE's prepackaged solutions are available

through an extensive partner network, and cover horizontal applications, vertical industry offerings, and application

development environments, allowing customers to focus on business innovation instead of integrating, validating and

managing IT infrastructure.

For more information, go to www.vce.com.

THE INFORMATION IN THIS PUBLICATION IS PROVIDED "AS IS." VCE MAKES NO REPRESENTATIONS

OR WARRANTIES OF ANY KIND WITH RESPECT TO THE INFORMATION IN THIS PUBLICATION, AND

SPECIFICALLY DISCLAIMS IMPLIED WARRANTIES OR MECHANTABILITY OR FITNESS FOR APARTICULAR PURPOSE.

Copyright 2011 VCE Company, LLC. All ri ghts reserved. Vblock and the VCE logo are registered trademarks or trademarks of VCE Company,LLC. and/or its affiliates in the United States or other countries. All other trademarks used herein are the property of their respective owners.

Vce Exchange Solution Architecture

Documents

Transcript of Vce Exchange Solution Architecture