Memory leak compressing ubuntu-17.04-desktop-amd64.iso

  • Alpha Testers

    I have read at that FreeArc Next has the better compression ration while compressing ubuntu-17.04-desktop-amd64.iso.

    So I tried the same with PA 17.00.69. Good news is that you guys, made it, making an even smaller resulting file.

    The thing is that I guess I have detected a memory leak, because memory usage reached the 2GB while compressing, which should not be the case, considering that a 24MB PLZMA dictionary with 4 execution threads is about 900MB of memory usage.

    For your information, in the same scenario, WinRAR was at 800MB of memory usage on peak.

    Can you please take a look? Maybe I am wrong, and this is the normal behaviour. Maybe Windows Task Manager is not properly reporting the usage.


    2GB options
    2GB advanced options
    2GB Ubuntu

  • conexware

    @nikkho there is no memory leak, those settings only apply to custom mode, not to optimized strong mode. So you are using a lot of memory of course, 64M at those settings (extreme), and also real 4 threads are used for that file. Advanced tab is least worked on feature in PA so far :). Many improvements there are still possible.

    That file is compressed, so i think advantage comes due to lzmarec coder + rep1 dedup. I think it would do even better with 128 MB, lets try it :)

  • conexware

    @nikkho btw, for me, real winner is Optimized Fast and Ultra or Max mode. It will finish in 30s on my computer and still be reasonably smaller due to rep1 :). Effect is better on other isos that are not as compressed as this one.

  • Alpha Testers

    With the new settings, the reduction was amazing, going from 1.571.555.291 bytes to bytes!!
    Switches wers: c_out=plzma4:mt4:c16M:d128M:mc16:fb273:lc1:lp0:pb0:a1; c_adr=plzma4:mt1:d22:lc0:pb2:lp2:fb273; c_hif=lzma2:x1:d22:mt1; rep1:fb15000:c300M:mem2000M { reflate:x9{ x64flt2{ deltb:c=1000K { deltb:c=1001K { c_out }} } / c_hif } }

    As espected, peak memory usage raised to about 6GB with those new settings, but speed was comparable to previous trial.

    Max 01

    Max 02

  • conexware

    @nikkho try c=256M, it will go down more :)

  • conexware

    @nikkho memory is not a problem on x64, it works well. Of course, it is also dependent on dictionary size.

  • conexware

    of course, for us, we cant set more than 64m by default, it would be unreasonableā€¦ but sure, it is fun for testing :)

  • conexware

    check with large D with new release:

Log in to reply