Building Capacity Mgmt Software Requirements

[Michelle Barton]

I'm in the process of developing requirements for the purchase of Capacity Mgmt Software. I manage windows, hpux, linux, and vmware. If you have these environments, what are the tools you use that you couldn't live without (functionality not vendor specific)? What do you wish you could do? What requirements did you have if/when software was purchased?

Thanks,
Michelle

[Carl Kyonka]

Michelle,

Our environments include Windows, HP-UX, Solaris and VMWare. We have some Linux lurking under desks, but nothing formal. The most basic type of tool is something to collect performance data. If you plan to do modeling, some monitoring tool must collect modeling data, which can be more detailed (you could have two products and deploy as is cost/effective).

Another essential tool is something to do basic forecasting, maybe linear regression. Nice if the monitoring tool includes this, otherwise you need a monitor capable of exporting data. I think time series might be more applicable to this sort of forecasting so that would be desirable. A good stats package would do.

If you have an environment of any size (say > 100 servers?) and you have big applications, then I would say you need a load testing tool. Load tests can take a lot of work to set up, but the alternatives are to test in production or use a modeling tool. Load tests are more directly applicable provided the test environment and load represent production. Models are more widely applicable and let you try configurations you do not yet have in place.

If you can afford it, having both a load tester and a modeling package would be best. If you have to choose, I think the load tester is an acid test and the first one to get. (I know I am speaking in the hall of analytical modeling, simalytics and software physics. I might be convinced to change this first choice.)

We do not have a CMDB (Configuration Management DataBase), but it would be useful. Without it, you need to keep track of key configuration details like CPU model (unless the monitor does this). And there may be many configuration details that are significant to performance.

My $.02,
Carl

[nilesh gajjar]

Michelle

To do Capacity Mgmt/ Planning you need to two set of information (a) Performance Utilization (% CPU, I/O, RAM, N/w etc.) (b) Workload Details (trnx/sec), you need a tool that can use this information to forecast/predict the capacity requirements for future growth in workload. In summary you need a tool that can do the above three task or at least the third one (doing prediction based on current resource usage stats and workload details).

Not sure if there is a tool that can do all the three task mentioned above - if any one knows plz share your experience.

a) Performance Utilization - there are number of tools available that can capture the resource usage details, i.e Tivoli
b) Workload Details - this is bit tricky, as there are number of way in which you can gather this details. By analyzing the logs of web/app/db servers, or using a smart sniffer that captures the data from wire. The second approach works best in production environment.

-nilesh

[Tim.Browning]

These are very good responses. I would like to offer one more suggestion, and that is to consider multiple, inexpensive, and simple tools rather than one large and complex toolset. Integrate them into your solution also as simply as possible.

Some of the larger and more comprehensive products demand a long learning period for the capacity planner because of their complexity and the tasks related to data management, software administration and maintenance, etc.. This toolset adminstration activity can use too much time and effort. Many IT environments have people 100% dedicated to managing the tools rather than using them. For example, a "SAS/ITRM Administrator" may be an expert at installing and fixing and customizing SAS/ITRM, but not very skillful in using it or creating programming solutions with it.

If you're hired to bake a cake, you don't want to spend 80% of your time repairing and cleaning your fancy do-it-all, fragile, and complex digital programmable oven. Show me the cake.

Tim Browning
Kimberly Clark Corp
ITS Consultant - Capacity Planning & Performance
tim.browning@kcc.com

[tscottt]

When I've worked with clients to device CM practices, tools and processes, I've found that the "mix" of tools depends on the environment. If you have one or a small number of highly integrated tiers serving up some core business applications, then a dedicated modelling tool such as made by TeamQuest or BMC may be appropriate.

There are products on the market that can be used to shoe-horn application monitoring as well: Amberpoint (now owned by Oracle) and Correlsense for example.

If your organization has a heterogeneous mix, and a large enterprise, then you face a different set of problems. What I've seen clients do is get "best-in-breed" host monitoring software such as HP's OpenView, Microsoft's SCOM/MOM, or IBM's Tivoli, not to mention SAN performance data collection tools. The problem there is that clients end up with a huge amount of performance data, and a set of static alarms that help with performance monitoring - but they under-leverage the potential of the dataset for capacity planning. Also, these tools do a terrible job of dealing with huge datasets from thousands of entities in terms of spotting capacity-relevant (as opposed to real-time performance) issues. Usually, clients in this situation end up developing a witches' brew of custom analysis and reporting tools, using SAS, Excel, R, etc.

I've developed a software solution aimed at pulling together, analyzing and reporting on huge performance datasets for such clients - but I won't turn this into a marketing email.

Scott Thompson
scott.thompson@concordusa.com
Director - Capacity management
Concord
http://www.concordusa.com