Results 1 to 8 of 8
  1. #1

    Bandwidth questions with SilkPerformer

    I'm in the process of architecting two geographically separate test labs that will be connected by a dedicated point-to-point circuit (T1/T3). One question is how many virtual users can I expect to concurrently run on a T1 circuit? A T3 circuit? In the short term we want to test with 2000-5000 virtual users. There will be some PCAnywhere and ssh traffic as well, although this will be mostly nominal. Also, does anyone have experience load testing with burstable circuits? ($$$ is always a factor)

    Thanks in advance!

  2. #2
    Senior Member
    Join Date
    Dec 2000
    Portland, OR

    Re: Bandwidth questions with SilkPerformer

    I don't think I can really answer your question, but one point to think about is what are these users doing?

    Are sending a lot of traffic over the wire?
    Do they mainly sit around and wait for "user actions" (think time in most tool parlance).
    Are they loading a lot of graphics or just text?

    What type of traffic load (in kbytes/sec) are you expecting for one user? What size pike will you actually think you will be using?

    Now as far as network overhead for a virtual user, or for PCAnywhere, I am clueless on these ascpects, but you are on the right track in including this overhead in your computaions.

    Good luck!

    It Depends.

  3. #3
    Senior Member
    Join Date
    Feb 2001
    Arlington, VA, USA

    Re: Bandwidth questions with SilkPerformer

    The number of virtual users you mentioned should all fit into a T1 although with noticeable slowness.

    One thing to consider though is will these networks be dedicated subnets or will other office workers be attached to them as well?

    I alwayas recommend using dedicated submets to:

    A)get the most accurate results since it is sort of a clean system.
    B)you don't want to slow down the connections of the rest of the company.

    just some thoughts.


  4. #4

    Re: Bandwidth questions with SilkPerformer

    Depending on what your virtual users are doing, 2000 users may not fit in the bandwidth of 1 T1. Here are some things you want to think about:

    1) model your customer behavior -- what's the average size (kbytes) of a page on your site? Are there any other types of files that would use up bandwidth, ie., applets or other downloads? Then figure approximately how frequently these downloads occur, taking into account caching at the client. That should tell you how much bandwidth you need for the site simulations.

    2) tool overhead: command between controller and agent during script runs are minimal, so ignore this. Post run -- the number of files transfered (results files) from agents to controller shouldn't be a problem either but make sure that firewalls don't lock up or cut off agent/controller communication mid-stream. What isn't so trivial is using the monitor aspect in performer -- to monitor your servers during a load test. So far, segue has no figures on this that I know of, better to just run the monitor and ask your network team to measure tha bandwidth usage.

    3) what is the connection speed you will be simulating? what does your company expect? For instance, if you're simulating 14.4 modem speed, your simultaneous bandwidth usage will be much smaller than if you are simulating t1 usage.

    4) Don't trust performer where the network is concerned -- do your homework first. I have seen performer run a test "successfully" and generate results when all network access from agent to server was cut off. Performer had no idea that the bandwidth needed to actually succeed in generating load was gone. There were no errors generated and no users were halted. This is a big problem in performer. So, to be absolutely sure, use some of your network resources to run an snmp agent on the virtual user agent and actually measure the bandwidth used. If you're running too close, add agents to distribute the network usage, if the SUM adds up to more than the bandwidth you need, well, you'll know you need more.

    Burstable bandwidth -- once you've done the calculations above, it shouldn't be too hard to verify bandwidth is bursting smoothly. Your challenge will be to make sure your agents are using the bandwidth you think they are.

    Good Luck!!


  5. #5

    Re: Bandwidth questions with SilkPerformer

    You might consider doing this on a bench before hitting your actual t1/t3.

    Use two routers (pick your favorite vendor). Define a link between the two routers with appropriate link speed and noise characteristics. Run test.

    If things go awol it's a lot easier to fix it within one room than one city/state/country/continent.

    Ebay usually has some bargain bid prices on used routers and short racks. You're best to stick with the vendor flavor that you already have since you can possibly pull expertise from your NOC group to help setup.

    Once you tune it in the lab, you can suggest your changes to router configurations (if needed) prior to running the app load test.

    You also might consider a commercial network load testing tool, such as Chariot from NetIQ, to pre-qualify an existing network infrastructure for numbers of transactions prior to layering an app load test on top of the network infrastructure.

    Under full disclosure, I do not sell Chariot, but I do sell services for training and consulting for Chariot.

    best regards,

    James Pulley
    iTest Solutions, Inc
    704-243-2854 (voice)
    James Pulley

    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

  6. #6

    Re: Bandwidth questions with SilkPerformer

    One thing you could do if you want a more realistic test is to dual-NIC your target Servers and Load-Generating Servers (providing they aren't already utilizing two NICs) and create a "control" network and a "test Network". This way, all the tool overhead and monitoring overhead is on one network leaving only the real test bandwidth on the other load.


  7. #7

    Re: Bandwidth questions with SilkPerformer

    Thanx for the reply.I am a bit new to this corelation stuff.I did try that out.I gave a rightboundry and a left boundry of the value it has to capture.The value that is captured is stored in a system defined variable "WCSparam".Now how do i go ahead.Do i have to substitute all the future occrances of that value by this parameter how do i go ahead.plz reply

    still confused...


  8. #8
    Senior Member
    Join Date
    Nov 2001
    Wilmington, NC, USA

    Re: Bandwidth questions with SilkPerformer

    I have come up with an average bandwidth usage of 166kb per user. But this is really going to depend on what you are actually doing.




Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
BetaSoft Inc.
All times are GMT -8. The time now is 05:52 PM.

Copyright BetaSoft Inc.