Plan Ahead with System Center Capacity Planner
At a Glance:
- SCCP basics
- Creating a new model
- Fine-tuning the model
- Simulating actual usage
Today’s infrastructure components are critical to the businesses they support. However, as the number of users increases and companies become more diverse, accurately estimating the size and
number of servers to run the components on becomes a challenge. Insufficient hardware leads to poor response times for end users and too much hardware is simply a waste of budget and entails additional management. For this reason Microsoft created System Center Capacity Planner (SCCP) 2006.
At a high level, SCCP allows you to enter information about network topology (locations, link speeds, latency), software and hardware to be used, and usage profiles (such as number of messages sent). SCCP then runs simulations of different usage levels and details the utilization of hardware and network components. This modeling helps ascertain whether the proposed configuration would be adequate in meeting your service requirements if implemented.
SCCP does not need to run on any specific server or perform network testing; it is essentially a simulation tool that can run test scenarios against a modeled environment. The actual system requirements for SCCP are minimal: Windows® XP with Service Pack 2 (SP2), Windows Server® 2003 or Windows 2000 with SP4, plus Microsoft® .NET Framework 2.0. The next generation of SCCP can be expected to interface more closely with Systems Management Server (SMS) and Microsoft Operations Manager (MOM), and to automate some of the information gathering since MOM already collects information about server utilization and SMS maintains information about hardware configuration.
The installation of the SCCP 2006 is simple; a standard installation wizard only requires license agreement acceptance and confirmation of the installation folder. Once the product is installed, a shortcut will be added to the All Programs menu. When you start SCCP for the first time, the Microsoft Capacity Planner Welcome screen provides access to three of the four components of the capacity planner as shown in Figure 1.
Figure 1 Starting the capacity planner tool (Click the image for a larger view)
The Model Wizard asks you questions about the desired environment in order to create a System Architecture Model (SAM). The SAM includes detailed locations, users, network, usage patterns and hardware.
The Model Editor allows for fine-tuning and modifications to SAMs previously created via the wizard or editor. This is where you can validate models, display simulated usage, and test other scenarios such as additional load on mail servers or the presence of remote users.
The capacity planner comes with a number of hardware types predefined for use in SAMs. However, if your needs differ, the Hardware Editor supports custom definition of hardware configurations.
When a model is selected, the Capacity Planner Simulation allows various tests to be performed against the selected model in order to determine its viability for actual real-world implementation.
The best way to look at the components and how they operate with each other is to actually use the tool. I’ll start with the wizard, run a simulation, and then fine-tune an architecture model by adding custom hardware. In this walkthrough, I will deal with an Exchange environment, but the process works in a similar manner for MOM.
Creating a New Model
Before you can perform validation or "what-if" scenarios on a planned infrastructure, or even if you just want to run scenarios about your currently deployed infrastructure, the environment must be described to SCCP. Unfortunately, you cannot simply scan the network in the current version. Instead, you must detail your infrastructure in terms of physical locations and types of connectivity between them, placement of servers, mailboxes, and users, and details about how the users will employ the environment.
There are a few limitations to the tool, which are explained in the user guide in detail. For example, in an Exchange model, only a single forest is supported. This means domains in external forests are not modeled, and any front-end servers are assumed to have both a Global Catalog and Mailbox store in the same site. Only 100 sites can be modeled of which up to 10 can be hub sites, a maximum of 300 server computers, and only Outlook® 2003 clients (Web or local) are supported. The good news is that SCCP supports up to 49,999 clients in central/hub locations, and up to 10,000 in branch locations. Exchange features such as ActiveSync® and Outlook Mobile Access connections are not supported, nor are legacy connectivity options such as POP and IMAP. Utilization caused by backup and restore operations are not accounted for and neither are cluster configurations; however, the capacity planner has the required options to accurately model typical day-to-day mail usage.
To begin the walkthrough from the Welcome page, select "Create a model with the Model Wizard" and describe the infrastructure to be modeled. I’m going to create a company that has three locations—a central location in Dallas and branch offices in Houston and Austin—each of which has a number of local e-mail users. Each location does not necessarily have its own servers, so I provide information on the number and type of e-mail users; SCCP will calculate in which locations local Exchange servers should exist and where users will access their Exchange server remotely.
After creating a basic model of the enterprise infrastructure, the wizard asks for the number of central and branch sites, the connectivity between sites, and the percentage of bandwidth that can be used for messaging purposes. The wizard takes a broad approach to get things running; later on I can fine-tune the model created by the wizard to configure individual links with more granularity. In this example, I can confirm the one central office and two branch sites (see Figure 2). There are T1 connections between the branches and the central site and half the bandwidth can be used for messaging. I do not need to configure the speed between central sites since I only have one.
Figure 2 Configuring WAN topology and connection bandwidth (Click the image for a larger view)
The next window of the wizard asks for the average number of users at the branch locations, the e-mail client (Outlook 2003 in cached mode or Outlook Web Access), and the number of users at the central site. Again this is very high-level information and can be fine-tuned later using the Model Editor. In this case I’ll say that each branch has 200 users and the central office has 500, all of which are using Outlook in cached mode.
The wizard has three default mail usage levels: High, Medium and Low. Each level assumes a different average mailbox size, number of messages sent and received, and average size per e-mail message. If none of the built-in types match the usage patterns expected in your organization, individual values can be configured for each of four separate usage characteristics.
The next step is to configure the hardware preferences as shown in Figure 3. SCCP allows up to three possible CPU configurations to be specified, and these will be used to model hardware for the various server roles. CPU configurations range from a single processor 1 GHz Pentium III Xeon all the way up to quad processor 3.66 GHz Xeon MP. Exchange 2003 consists of a number of roles, namely Mailbox Store, Bridgehead, and Front End, plus an additional Active Directory® Global Catalog role. If utilization permits, the option is available to consolidate roles onto fewer servers. If server consolidation is not specified, then separate servers will be configured for each of the roles even if utilization does not require it. Allowing SCCP to consolidate will provide the best results.
Figure 3 Configuring hardware profiles for servers (Click the image for a larger view)
Finally, the type of disks available can be selected based on the interface, size, and speed. The number of disks is not required; the capacity planner will calculate it for me. The planner can also specify a Storage Area Network (SAN), although the ability to create clusters is not currently available.
The next step in the wizard is a summary of the calculated model, which shows the number of servers SCCP calculates as optimum, the roles of each of the servers, and details such as processor, memory, and disk configurations. If a SAN had been selected, it would also be displayed.
Note that just because I selected multiple locations, it is quite possible that servers are only displayed for a subset of the locations. This is because SCCP ascertained that local Exchange servers at some locations are not optimal and instead mailboxes are better served via a remote Exchange mailbox server which is the case in my example as shown in Figure 4. In this case, the 400 branch office users have no local Exchange mailbox store, but instead have been consolidated onto the single Exchange mailbox store server in the Dallas location which now shows 500 users.
Figure 4 A summary of the initial capacity planning evaluation (Click the image for a larger view)
Up to this point in the wizard I can click the Back and Next buttons to modify answers and fine-tune the model summary (or directly click on wizard pages via the left-hand navigation pane). When satisfied with the configuration, I click the Finish button on the Model Summary page. This opens the model in the Model Editor, which is where more granular configuration is possible.
Fine-Tuning the Model
The Model Wizard is a great tool for creating a foundation model. However, unless the various locations are all the same and averaged out, some customization will likely be required for the model to be truly useful as a basis for calculating utilization. It’s not unusual for locations to have different numbers of users, or perhaps a combination of the various messaging clients such as some Outlook 2003 cached and some Outlook Web Access (OWA). While it’s not possible to get to this level of detail in the Model Wizard, it is configurable via the Model Editor. This tool is also useful when trying to ascertain the effect of modifying aspects of the environment, for example by adding additional locations or types of users.
The Model Editor, shown in Figure 5, saves and retrieves models in .sam files, which are a proprietary format to the tool. It is good practice, as with all work, to save models to file before major changes are made, and to use a descriptive name to ensure no working models are lost. The File menu provides an Export menu that generates a detailed summary report in Excel® format and as a Visio® topology diagram of the model.
Figure 5 Tweaking a SAM configuration in the Model Editor (Click the image for a larger view)
In the left-hand pane of the Model Editor is a basic navigation area with options to see the global topology, site topology, or a model summary. Global Topology shows the architecture model from a site perspective with additional networks such as the Internet. The Site Topology view displays information about a selected site or network. Model Summary displays a textual summary of the environment, which is the same format as what you’ll see displayed at the end of the Model Wizard.
Whichever option is selected, the details are displayed in the center pane of the Model Editor. The right-hand pane—in the spirit of the new version of the Microsoft Management Console (MMC) that shipped with Windows Server 2003 R2—is an "actions" area that is split into two sub-areas: the top part provides context-sensitive actions that change depending on the current view or object selected, while the bottom part contains general actions that always apply. These latter actions let me add a new site, edit usage profiles of users, and (the fun bit) run a simulation.
When working in the Model Editor center pane, all of the items are selectable. After selecting an item, in addition to the current item actions changing based on the selected object, I can drill down into more detail about the object. For example, double-clicking on a site will switch to the Site Topology view and give details of the objects modeled for that site. Thus, a locally served site would show one or more servers, networks and the users at the location. This makes it possible, in turn, to double-click the selected object to view its detailed properties of and then modify them as needed.
The lines that connect the sites represent the networks between the locations and can be edited to allow exact configuration of the network. For example, I can specify different uplink and downlink bandwidth speeds and the percentage of the bandwidth that is available. I can also configure connection latency, which can be a very important attribute for stable communication. Other common changes include adding additional user types. In this instance, let’s say that I have a mixture of Outlook Web Access and Outlook 2003 clients serviced by the central location. After opening Dallas in its Site Topology view, I would select "Add a client profile" and add information for the OWA-based users that would include the server that the OWA clients will use, the network the users will connect from and the bandwidth that will be available to them, as shown in Figure 6.
Figure 6 Adding and modifying elements in a model is easy (Click the image for a larger view)
Once the SAM reflects the existing or planned infrastructure as closely as possible, save and name it accordingly to ensure that it is easily identifiable as the starting point for any future modifications. At this point, I like to run a simulation to get a benchmark of how the model will perform under the configured settings. Based on that benchmark I can modify the model to test various "what if" scenarios such as adding a new site and users, adding or removing servers, adding new users to locations, or changing the type of connection or e-mail client. Once the model has been updated with these new elements, I can run another simulation to see how the new configuration holds up.
Simulating Actual Usage
Running the simulation is as simple as clicking the Run simulation general action or toolbar button. A sanity check of the model happens first: any flaws in the model are displayed before the simulation is performed. For example, if a number of OWA clients are added to a model that previously had no OWA clients and no Front End server was configured, an error message will notify you that no Front End server exists. If no such flaws exist, status will be displayed as the simulation is performed.
Once the simulation is complete, a report with detailed information about the usage of all components is generated and the Simulation Results view will open. It initially displays a summary view giving details of highest CPU, storage and connection link utilizations (which are common bottleneck causes) as well as longest transaction and latency times, which may affect end-user experience. Details about exact utilization of the servers, SAN, local and external connections can be viewed by selecting links in the left-hand pane of the simulation results as shown in Figure 7.
Figure 7 Simulation results summary for a simple model (Click the image for a larger view)
Holding the mouse over the various graphical displays, such as the Bottleneck analysis, shows the server and the attribute of the server that is causing that bottleneck. Notice the Threshold settings in the left-hand pane. This category allows me to set what I consider reasonable usage before raising warnings, a useful feature.
The detailed simulation reports are great not just for helping to inform hardware purchase decisions but also to show the amount of network resources that will be used by the messaging applications. For example, for users connecting via slower WAN links, the report provides an indication of whether the link is fast enough. If it’s not, I can find out if the speed of the link between locations should be upgraded or if a local mailbox store server should be added, simply by modifying the model and re-running the simulation to get results that fit specific criteria. Likewise, if the servers are all running at 5 percent CPU utilization, I can drop the specification of the servers. By allowing for easy model modification and the ability to re-run simulations, I’m actually free to experiment with different solution options. With accurate metrics from all of the tested scenarios, I can walk into budget and planning meetings with confidence.
I just gave you an overview of how SCCP enables you to simulate and analyze Exchange deployment. Keep in mind that SCCP performs the same level of modeling and simulation for MOM as well. SCCP can currently be downloaded as part of MSDN® Premium downloads or TechNet Plus. ¦
John Savill is Director of Technical Infrastructure for Geniant. He is a CISSP, a Security and Messaging MCSE on Windows Server 2003, an eight-time MVP, and a Krav Maga instructor. He is also the author of Windows Server 2003 Active Directory Design and Implementation from Packt Publishing.
© 2008 Microsoft Corporation and CMP Media, LLC. All rights reserved; reproduction in part or in whole without permission is prohibited.