Projects/Utils/kwallet/Benchmark

    From KDE TechBase
    < Projects‎ | Utils‎ | kwallet

    Summary

    One of the main concerns users had with the wallet was that new entries weren't saved immediately (bug #105752). Before actually getting to work and coding a workaround (appending new passwords to the file without reencrypting all of it or creating a second file where new passwords are appended) I benchmarked the existing encryption to check the actual overhead that would be incurred by saving to the kwl file as soon as passwords were entered.

    Code

    I used the following code for benchmarking. Please keep in mind:

    • keys and passwords generated using random data are probably longer than the entries you actually have in your wallet.
    • only syncing the wallet is benchmarked
    • due to hd caching (and a rather modest filesize), most of the time reported should be used encrypting the data.
    #include <kaboutdata.h>
    #include <kcomponentdata.h>
    #include <kcmdlineargs.h>
    #include <kdebug.h>
    #include <kwallet.h>
    #include <QApplication>
    #include <QTime>
    #include <QFile>
    
    #include "../backend/kwalletentry.h"
    #include "../backend/kwalletbackend.h"
    
    using namespace KWallet;
    
    static int getRandomBlock(QByteArray& randBlock) {
      QFile devrand("/dev/urandom");
      if (devrand.open(QIODevice::ReadOnly)) {
        int rc = devrand.read(randBlock.data(), randBlock.size());
        if (rc != randBlock.size()) {
          return -3;              // not enough data read
        }
        return 0;
      }
    
      return -1;
    }
    
    
    int main(int argc, char **argv) {
      KAboutData aboutData("kwalletbench", 0, ki18n("kwalletbench"), "version");
      KComponentData componentData(&aboutData);
      QApplication app( argc, argv );
    
      Backend back("/tmp/benchmark.kwl", true);
    
      back.open("benchmark");
      back.createFolder("benchmark");
      back.setFolder("benchmark");
    
      for (int i = 0; i <= 100000; i+=100) {
        for (int j = 0; j < 100; ++j) {
          Entry entry;
          QByteArray key, value;
          key.resize(20);
          if (getRandomBlock(key) != 0) {
            kDebug(0) << "Error";
          }
          value.resize(50);
          if (getRandomBlock(value) != 0) {
            kDebug(0) << "Error";
          }
          entry.setType(Wallet::Password);
          entry.setKey(key);
          entry.setValue(value);
          back.writeEntry(&entry);
        }
    
        QTime _start = QTime::currentTime();
        back.sync("benchmark");
        QTime _end = QTime::currentTime();
    
        kDebug(0) << i << ";" << _start.msecsTo(_end);
      }
    
      back.close();
    
      return 0;
    }
    

    Results

    I benchmarked on a Q6600. As the encrypting code is single-threaded, only one core (2.4GHz) is being used. Please bear in mind that I didn't bother to run the test several times as the results are pretty clear - unfortunately this makes some of the numbers seem a little weird.

    • DebugFull
      • 1 password: 16ms
      • 100 passwords: 15ms
      • 1000 passwords: 32ms
      • 5000 passwords: 107ms
      • 10000 passwords: 192ms
    • Release
      • 1 password: 8ms
      • 100 passwords: 5ms
      • 1000 passwords: 15ms
      • 5000 passwords: 25ms
      • 10000 passwords: 56ms

    Discussion

    I assume that your usual wallet will contain less than 1000 entries. Due to the fact that any workaround would have to encrypt at least 1 entry (~ 8ms) this workaround would save around 15ms - 8ms = 7ms. This is clearly insignificant. Big overhead for syncing the wallet seems to be a myth.