Performance of Top 50 Websites via Tor

Discussion in 'privacy technology' started by mirimir, Jun 16, 2014.

Thread Status:
Not open for further replies.
  1. mirimir

    mirimir Registered Member

    Joined:
    Oct 1, 2011
    Posts:
    6,025
    I've decided to test the performance of the top 50 websites (according to Alexa) via Tor. I'm using a script that runs "torsocks midori -s http://foo.bar/" to snapshot sites. Midori is configured send a Firefox user agent string.

    Results from the initial test series are [here][1]. I can't upload the file to Wilders because it's too large. I'm also posting results to Tor.SE[2]. Some exemplary histograms are [here][3]. It's interesting that microsoft.com and craigslist.com have such similar bimodal speed distributions. And it's amusing that a porn site has the highest speed among the top 50 ;)

    I'll add additional text, highlighting the results, as I have time. I'll also update the image link as I include new data in the analysis.

    [1] http://www.4shared.com/download/LR2cDKL-ce/Top_50_Websites_via_Tor_rev.png?lgfp=3000
    [2] https://tor.stackexchange.com/quest...sites-are-known-to-be-troublesome-for-tor-use
    [3] http://www.4shared.com/download/QphEYo8Jba/Histograms_for_Example_Website.png?lgfp=3000
     
    Last edited: Jun 17, 2014
  2. Tipsy

    Tipsy Registered Member

    Joined:
    Aug 25, 2013
    Posts:
    207
    What is the download speed measures? Just time for download the main page, such like to download the main page to www.google.com?

    Is correct that you are finding www.yahoo.com loads almost 10 times faster than www.google.com through Tor?
     
  3. mirimir

    mirimir Registered Member

    Joined:
    Oct 1, 2011
    Posts:
    6,025
    I calculated download speed by dividing file size by time to download. The testing script logs the current time just before starting the midori snapshot. The timestamp on the snapshot is the completion time. And so time to download is timestamp on the snapshot minus starting time.
    No, google.com loads in about 10 seconds on average, whereas yahoo.com loads in about 14 seconds on average. However, the average size of google.com is about 38 KB versus 416 KB for yahoo.com. Given the size difference, google.com loads at about 37 Kbps, whereas yahoo.com loads at 305 Kbps.

    There's another factor. For small pages, circuit latency accounts for much of the load time. And so throughput is arguably much larger than the download speeds that I've calculated.

    For similarly sized pages, the download speeds are arguably comparable.
     
  4. Tipsy

    Tipsy Registered Member

    Joined:
    Aug 25, 2013
    Posts:
    207
    OK, I see.
    "For similarly sized pages, the download speeds are arguably comparable." That is an important point that maybe is not so clear in your table.
     
  5. mirimir

    mirimir Registered Member

    Joined:
    Oct 1, 2011
    Posts:
    6,025
    Good point :)

    I haven't even started a serious writeup yet.

    I'm mainly looking at relative loading speed for each site, as a function of the Tor exit used. Some sites (e.g., microsoft.com, adcash.com and craigslist.org) sometimes load quickly via Tor, and sometimes load ~1% as fast (with the same page size). See the histogram chart.
     
  6. Veeshush

    Veeshush Registered Member

    Joined:
    Mar 16, 2014
    Posts:
    643
    I remember the days when Tor almost felt like dialup. Now I can't honestly tell a difference less I'm actually downloading a large file (like 400mb + ).
     
  7. mirimir

    mirimir Registered Member

    Joined:
    Oct 1, 2011
    Posts:
    6,025
    Hit craigslist.org a few times ;)
     
  8. mirimir

    mirimir Registered Member

    Joined:
    Oct 1, 2011
    Posts:
    6,025
    I've been thinking more about this issue. Some simple pages (e.g., google.com) load very quickly because they're small. Yet some larger pages (e.g., yahoo.com) load as quickly. And so calculated throughput can be much greater for those larger pages. However, it seems doubtful that yahoo.com (for example) has over 10 fold bigger pipes than google.com does ;) But maybe google.com handles far more users per server.

    Anyway, I'm wondering whether I can adjust calculated throughput based on latency. But ping doesn't work via Tor :( However, netcat does work:
    Code:
    date +'%F %H:%M:%S.%N' ; torsocks nc -z foo.bar 80 ; date +'%F %H:%M:%S.%N'
    Does that seem an accurate measure of connection latency?
     
  9. Tipsy

    Tipsy Registered Member

    Joined:
    Aug 25, 2013
    Posts:
    207
    Yes, that seem like good idea. But somebody more experienced than me should say.
     
  10. caspian

    caspian Registered Member

    Joined:
    Jun 17, 2007
    Posts:
    2,301
    Location:
    Oz
    I have recently started using Tor on a message board. In the past I would have never considered this. But it is almost shocking to me how much faster Tor is these days. It's actually pretty darn usable now. I always have a VPN connected though in case it leaks, since I have to allows scripts on certain sites. But it has been working beautifully on the message board that I frequent. I don't use it on Google or Yahoo though because they greak out. They think I am being hacked or something.
     
Loading...
Thread Status:
Not open for further replies.