Calculation of Vuser per Load Generator
1. What version of LoadRunner (LR) or PerformanceCenter (PC) are you using? (specify which tool LR or PC)
(We are using LoadRunner V 9.1)
2. What is the protocol you are recording?
(HTTP/Web, SAP GUI, SAP Portal, Citrix)
2.1 If .NET, which version of .NET does the app require?
2.2 If .NET, which version of .NET do you have installed?
2.3 If OracleNCA with Oracle Forms Server, please list the version of Oracle Forms Server. You can get this info from the main Oracle Form when launched - Help -> About
3. If HTML - are you using HTML-Advanced with URLs or,
4. If URL mode:
- Concurrent groups, or
- Without Concurrent groups?
5. Which LoadRunner/PerformanceCenter feature (FPs) or service packs are you using?
6. VuGen Recording - are you using Old or New Recording Engine?
7. You must list here the specific Licensed Vuser type for your specific issue AND the license amount you have for your Vuser protocol(s) - per the example below (Unlimited, Permanent, N/A, etc. are not options) If you do not know because you are using a cracked version, list as "CRACK".
HP LoadRunner SOA Protocol Bundle 2500 Virtual Users SW LTU
HP LoadRunner SAP Protocol Bundle 2500 Virtual Users SW LTU
HP LoadRunner Web Multimedia Protocol Bundle 2500 Virtual Users SW LTU
HP LoadRunner Remote Access Protocol Bundle 2500 Virtual Users SW LTU
8. Is your support/maintenance contract current and active?
( Yes )
9. What platform(s) (PCs) and Operating Systems (Windows-XP, etc.) are being used for load generators and controllers? Include version and service packs (SP1 or 2, etc.)
Load Gens' Windows XP SP2
Controller VMware 3.5
10. If you have filed a service request with HP/Mercury, what have they told you at this point with respect to your issue?
I recall a post on this site that outlined a procedure for calculating the number of users that could be run on a load generator (using HP Load Runner v 9.1). Is it possible someone has that post and could direct me to it.
I know we measure amount of system resources (e.g. Processor, memory, disk,) used by the application under test.
We are using HTTP/Web and SAP GUI protocols.
I am setting up my Load Generator environment to begin on a SAP deployment. Eventurally we will be testing a user base in excess of 50,000 users.
Re: Calculation of Vuser per Load Generator
Search for posts by James Pulley - this sounds like information he might have posted.
For further assistance please read and follow:
Re: Calculation of Vuser per Load Generator
User base of 50K, using standard corporate model spread of concurrent and active (5-12%) places your user session count in the range of 2500 to 6000 concurrent users
User base of 50K, using internet standard model spread of concurrent and active (.5-2%) places your user session count in the range of 250 to 1000 concurrent users.
If you were looking at active and concurrent number of 50K users then you would be looking at user model populations of 1,000,000-416,667 (corporate) or 10,000,000-2,500,000 (internet standard model facing.)
You look like your license is falling within the corporate model range for a user population of 50K between 5-12% concurrent and active.
As to Load Generators, there is capacity question and then there is best practice. If you are testing an application with that much risk, go ahead and follow the best practices. The cost of a generator population is going to be a gnat's eyelash as compared to the cost of your LoadRunner license.
Here are your rules of thumb:
<ul type="square">[*]No Virtual User Load on the LoadRunner controller host[*]Not less than three generators (primary load on two plus a single virtual user of each type on the third)[*]Always have a hot spare, hardware matched, to use as a cannibal for failures or to press into service to rebalance load.[*]All hardware matched, max memory (4GB for 32 bit CPUs and OS's - Load Generators are 32 bit animals), dual or quad core at this point, 200GB HD min.[*]Windows XP Sp3 on generators. You don't want the overhead of Server or Vista to eat into your CPU or Memory resource.
[Intelligent network adapters, get them with TCP offload engines and SSL accelerators if possible. This takes a lot of interrupt service load off of the core CPU. You can extend this analogy to the video cards and the disk subsystem as well. Get co processed video and intelligent disk adapters, with the goal of leaving the core CPU to the task of running the OS and the MDRV agents[*]Run with blank screen saver on all load generators[/list]
I always recommend a client purchase at least seven hardware matched solutions, accounting for the one in seven hardware failures in a given year. Take on for the controller, a second for the "control" generator, a third for the hot spare, which leaves four for your primary load.
Given your projected load in the 2500-10000 (if you run all virtual users in one test) range, you might consider more machines. I tend to be very conservative with my virtual users as I never want to have a capacity issue in my test equipment color the test results and then have to run the test over. If you looked at a 10K test, with 500 virtual users per box (you could probably go to 1000-1500 depending upon the virtual user without issue), then you are looking at 20 boxes for primary load (+1 Controller, +1 Hot Spare, + 1 control). Even if they were 1K apiece, which is extremely pessimestic, you are looking at a very small amount versus the cost of your LoadRunner license, the labor and the expertise to make this test execute correctly.
Could you get 1K, 2K or 5K on one host? Potentially yes depending upon the size and complexity of the virtual users. However history suggests that you will have a substantial number of issues related to your decision from a process, a test results integrity and results defensibility basis. Plus lacking a control factor is generally considered not best practice for testing whether you are testing a fuel injection system or a complex piece of software.
Replace ineffective offshore contracts, LoadRunnerByTheHour
. Starting @ $19.95/hr USD.
Put us to the test, skilled expertise is less expensive than you might imagine.
Twitter: @LoadRunnerBTH @PerfBytes