• 0

    posted a message on PitBull 4.0
    Quote from Shefki
    Like I said the test isn't perfect I didn't have time to duplicate your default config in raid frames in PB4. It's not easy making a fair comparison. So I kept it simple. Are you seriously complaining about a comparison that you won? Generally I tune over an entire AV run just like you said, my testing of various improvements are done that way. But it's far easier to make a tweak and run some AV's and see how it does that it is to setup two different unit frames to run at the same time and run some AV's. Sure I could drop both in with roughly default configurations. But that's a pretty ridiculous comparison when we've got a number of things you don't (which is perfectly fine).

    The data is accurate. You may think the tests were not the best that could be done, but I even admitted as much in my post. Facts are just facts. There's no such thing as a bad fact. It's either true or not. So are you suggesting that I just made up those screenshots?

    I'm really not seeing how I'm preptuating bad on ideas on how to measure performance. I'm looking at CPU time in the same environment. While running entirely disembedded.


    There's a difference between a test that might be a little bit off and a test that is invalid, yours is the latter. I care more about the accuracy of tests especially when they are posted publicly in a thread that users will read and get confused by.

    I am telling you that your method of profiling is incorrect, obviously you did not make up the screen shots but the method you used to gather the data was incorrect and so the actual statics you got should not be used.

    Stop mincing words, the data is accurate in the context that the numbers were correctly totally up by the mod using Blizzards API's. The data is not accurate in the context that you're trying to use them, which is that the performance difference between PitBull and Shadowed Unit Frames is only 35%.

    And stop taking this so personally just because I'm telling you your statics were not gathered correctly. It does not mean it's a personal insult to you. I care more about accuracy and the truth than your personal feelings towards this.

    If you want to argue the performance numbers are that different based on modules fine, but from the sounds of it you disabled the ones that weren't relevant so they shouldn't have been an issue in terms of CPU. Especially with the fact that I would assume you reset the CPU stats right before starting to profile to remove the startup counts.
    Posted in: Unit Frames
  • 0

    posted a message on PitBull 4.0
    Quote from Shefki
    So the matter of performance versus SUF has come up recently. I figured I'd do some testing to compare. First thing I did was install SUF and leave it configured in it's default state. Then I setup a PB4 profile that matched this as closely as I could. Features that PB4 has that SUF doesn't have at all had their modules disabled entirely. In general otherwise the frames were configured roughly the same. There were a few differences. I didn't bother to make the percentages be just integers like in SUF. Also SUF has configuration options for fast Mana updates on a per frame basis. While we turn off our functionality for fast updates when the CVAR is disabled we do fast Mana updates for all frames displaying the player.

    An idle test with the player targeted and focused. And a combat test with the mob targeted and focused. During the combat test I did nothing but melee the mob. All tests were one minute tests with the CPU profiling reset at the start.

    Screenshot of idle test results (note SUF is on top, PB4 is on the bottom):
    http://files.shefki.org/ScreenShot_062009_144400.png

    Screenshot of combat test results:
    http://files.shefki.org/ScreenShot_062009_145959.png

    All tests were done with LuaTexts. I don't think we're that terribly far off in performance. Our memory usage is much larger. At this time PB4 while it is modular still has not had the modules split out to be LoD. This means that even though the modules that weren't relevent were disabled, they still had their code loaded into memory and they still incurred a small amount of CPU time during updates to verify if they were Enabled or not.

    These tests are not perfect. But they were quick and dirty tests to give me an idea of where we stand. I didn't do raid frame tests, party tests or anything like that. Simply because I didn't want to spend the time to do so.

    I'll be happy to share the profile I used to do this testing with anyone that's interested.


    First off, idle tests are worthless. Anyone who optimizes code based on the user idling is over optimizing their code, I'm sure I could reduce the idle CPU if I really wanted at the cost of using more CPU while in combat (Very easy decision to make).

    PB4 is using 35% more CPU, that's not really something I would consider that close to SUF.

    Static memory is not a profiling stat. If someone is running a low end computer with only 512 MB or 1 GB of RAM, they might not be able to run as many addons or programs in the background (Really, they would save more memory running less programs in the background). The only memory statistic that can be used to profile with is the delta over a period of time.

    Oh and, your profiling stats are completely invalid because you tested it solo over 60 seconds and not over say an entire Alterac Valley match because raid frames and party frames are far far more depending than doing something solo for 60 seconds.

    I don't honestly care if you want to use profiling statics to back up a point, but gather your data accurately and stop using invalid statics. All your post is doing is perpetuating bad facts on how you can measure performance.
    Posted in: Unit Frames
  • To post a comment, please or register a new account.