Provisioning Methodology & Tools
Hi,
I'm looking for a supplier of methodology and/or tools to allow an IT organizatio to be a bit more "scientific" about estimating capacity requirements and provisioning infrastructure for new applications.
Ie we need help with the process to arrive at answers to questions such as:
What infrastructure elements do I need to design and provision so that application X has a guaranteed uptime of 99.999%, a subsecond transaction time with 10.000 simultaneous users, in a three tier architecture based on oracle, weblogic and apache, with f5 loadbalancing and checkpoint firewalls?
Software/Hardware used:
Software/Hardware used:
ASKED:
January 4, 2006 1:22 PM
UPDATED:
January 11, 2006 9:45 AM
Answer Wiki:
There are a few O-Reilly books on large scale networks and performance tuning. These are a good start. I worked a guy who did this for a living and there is no 'one way' to do this, however it always comes down to testing and measuring. N+M farms and Active/Active or Active/Standby type configurations are a given. Good analysis of where bottlenecks will occur and proving that is where to focus.
As for performance of number of users, you'll need the near complete app to test, or something that is equivalent..
To see all answers submitted to the Answer Wiki: View Answer History.
Hojans,
There is no way to accurately predict how any given infrastructure will cope with a new application…. there are too many variables. There are basically only two methods:
1. Best guess capacity plan which takes the things that are known, and the body of experience for each layer (the vendor’s best tuning advice for each component) and reasonable guesses for the unknowns, and allow a contingency in your budget in case it does perform in real life; and
2. Benchmarking. This involves setting up a test environment as close as possible to the planned target environment, and usually running a load generator of some kind (automated transaction load simulator). There are various ones available, depending on the environment. Talk to the infrastructure software vendors, they’ll know what runs on their environments.
Neither approach is bulletproof. Reading through the description above, you may think that benchmarking is more scientific…and it is, but it won’t guarantee a better result than the first option.
Hope this helps…
If you are buying for greenfield implementation, you could put the burden on your supplier, specially if you go with a single integrator / company.
As the previous messages states, get equipment to simulate the load and test, test test. In delivering telecom equipment in the past, everyone wants to see load testing done with several key statistical loads to determine the actual capacity. There are quite a few load generators around for web servers so you should have a reasonable range of choices.
best of luck,
Paul
To be what you termed “a bit more scientific,” as several other respondents mentioned you probably need to perform simulation. Further, a given application is not an “island,” so you probably need to build a model of your environment in order to run what-if scenarios, including the impact of new applications, equipment, etc. The objectives you cited – such as 99.999% up time – might already be out of reach because of service interruptions your firewall, DNS service, directory services, routers, etc. The model probably does not have to be greatly detailed, but its usefulness depends on having realistic experience data.