Which BitTorrent Client is the Fastest?
March 01, 08 by sharky 21,626 viewsFastest BitTorrent Clients - The Tests - A Comparison of Speeds in Realtime
Every BitTorrent software developer claims that theirs is the fastest BitTorrent client, but which one really IS the fastest? There are so many variables to be considered: Seed/peer ratio of individual torrents, public or private trackers (and number of trackers in a torrent), DHT “trackerless” support and UPnP port forwarding, just to name a few. Even the default “factory” settings of each client can have an big impact on download speeds.
So how it is truly possible to really know? Conduct speed tests in a controlled experiment, natch! Let’s put ‘em to a test!
About The Tests
To create a semi-controlled environment in which the data could realistically be used, we needed to conduct the experiment in realtime - in a scenario where ALL BitTorrent clients were using the exact same torrent (loaded simultaneously in each) and thus were ideally downloading at the same time. Not only that, but we repeated the same text three times with 3 different .torrents (all 5 clients were loaded with the exact same torrent at the same time, on three separate occasions).
OK, no peeking - don’t scroll down to the bottom just yet. Out of the five tested clients (µtorrent, BitComet, Azureus, BitTorrent and BitLord), which one do YOU think won the races?
The Variables
To cut down on some of the variables, we used the same torrent for each, from the same source (mininova.org) thus the trackers in the torrent would remain identical (and any changes to the tracker and seed/peers would be reflected identically for each in real time). We also “tweaked” the settings of each client to try and maximize speeds. Here’s what we attempted to do within each of the five BitTorrent clients, prior to running the torrent:
Set each to “maximum” download speed.- Set each upload speed to 50KB/s.
- Enabled DHT and UPnP where possible.
- Different port numbers (obviously) needed to be used, but we stayed within the good range for each client.
- Attempted to re-create the same ’settings’ environment through advanced connectivity options. Where applicable, each clients’ settings were modified to reflect these similar environments:
Although we conducted these tests using three different popular torrents, we only ran one torrent at a time for all of the clients. This ensured that we’d be able to achieve maximum or close-to max swarm speeds for each client during simultaneous downloading. The .torrent file itself is probably the most important constant of the experiment, which is why we chose to use the same torrent for all five (on three different occurences).
The Tests
We elected to use five different BitTorrent clients, and three different torrents to use in each. Also note that we (and so should you) have Peer Guardian 2 running during these tests. HTTP was allowed, and here are the blocklists we used:
The BitTorrent Clients
For the software, we wanted to provide the TOP 5 BitTorrent clients (in terms of usage and popularity). The first four in the roundup are no-brainers, but for the fifth one we relied on comparing alexa traffic rank data with number of downloads. In our opinion, BitLord rolls in at #5 (although this is debatable). Here’s the 5 clients we used in the testing:
- µtorrent (we used v1.6.1)
- BitTorrent (the original client) v6.0.2
- Azureus v3.0.4.2
- BitComet v0.96
- BitLord v.1.1
The .Torrent Files
For the *.torrents, we needed to supply an equal playing field for each client. To achieve this, we chose 3 different torrents from mininova.org (using public trackers). This would ensure that each client used the same publicly available torrent with the exact same trackers in each. Seed/peer statistics would thus be updated simultaneously for all clients.
We also opted for torrents that are proper “scene releases” listed on dupecheck websites (therefore eliminating the chances for anti-P2P tampering or amateurish uploads). Also note that all 3 torrents were one week old, in order to smooth out the highs-and-lows of seed/peer ratios normally associated with newly-released torrents. Here’s the 3 .torrents we chose:
- Rambo.IV.REAL.PROPER.R5.xVID-UNiVERSAL.4041898.TPB [mininova].torrent
- Sleuth.LIMITED.RERIP.DVDRip.XviD-DMT.[www.torrentfive.com] [mininova].torrent
- Bee.Movie.DVDRip.XviD-DiAMOND.4041897.TPB [mininova].torrent
The Results
Here are the results! All speeds are in KB/s.
| BitTorrent Client | Test #1 - Bee Movie | Test #2 - Rambo | Test #3 - Sleuth | Average D/L Speed: |
| µtorrent | 35.3 | 39.0 | 30.3 | 34.9 |
| BitTorrent | 31.9 | 30.9 | 25.4 | 29.4 |
| Azureus† | 44.0 | 39.1 | 28.8 | 37.3 |
| BitComet | 55.0 | 40.0 | 29.9 | 41.6 |
| BitLord | 51.0 | 39.0 | 31.0 | 40.3 |
†Azureus is the only client that doesn’t show a statistic for the “average” download speed of completed torrents. Instead, it shows data for the “time” it took to complete a torrent, in hours and minutes. To calculate the average download rate, for example - Bee Movie is 714.0 MB with a “Finished in” time of 4h 37m, we had to derive the KB of the file and divide it by seconds. 714 X 1024 = 731136 (total KB) divided by 4h 37m (277 minutes X 60) = 16620 seconds. Therefore the average download speed was 731136 / 16620 which equals 43.99 KB/s.
The winner? In our tests, BitComet is the fastest BitTorrent client. No surprise here, we could’ve already predicted that. It’s little wonder - BitComet has been banned by most private trackers for being a “dirty” client, and uses aggressive settings & features to maximize downloading.
NOTES: BitTorrent (the original Bram Cohen’s BT client) came in a dismal fifth place, and in all three torrent examples it was the last client that hadn’t finished downloading - to help it out, we removed the “seeding” torrents from all other clients so that it could finish downloading on it’s own with maximum bandwidth. If we hadn’t done this, BitTorrent’s speeds would have been even lower. µtorrent: We were surprised that it came in fourth place. Perhaps the version number may have had a factor - we used v1.6.1 (a private-tracker ‘friendly’ build number).
May this put your mind at ease!
01 • Blaenk Denum Says: 01.03.08 at 6:10 pm
I think DHT should have been off. There was no reason to keep it on and it just added to the variability of the tests. Also, BitComet shouldn’t have been included in the test due to it’s notorious reputation. It’s like saying that the winner is a client that cheats its way to the finish line, and what’s worse is you know it.
Like you noted, it is really difficult to benchmark any one client, for there are too many variables involved. I’m pretty sure you set the same options for most clients (Max number of connections, etc.), but there’s also the question of how many peers/seeds each client gets to connect to compared to any other client at any given time. Not to mention the bandwidth is being split four-times among each client (And that’s assuming you aren’t running anything that uses the network in the background or during the tests), and there could be a question of QoS regarding whether or not any one client hogs more bandwidth by marking itself with a high priority.
Anyways, you’re one of the few sites that really talks about these things in an informative manner, so keep the articles coming! I didn’t mean to come off as a nitpick, just thought I’d say these things anyways for any other readers to take into consideration (Rather than having everyone rushing to get BitComet). Perhaps a comparison of these clients would’ve been better. Regarding their system resource usage, their size, speed (In execution and possibly bandwidth), their features, options, and reputation (i.e. BitComet is banned by many trackers for ‘cheating’, etc.), platform support, and any other things that could be compared.
02 • sharky Says: 01.03.08 at 7:34 pm
Thanks for the feedback! That’s what this is all about - I’m thrilled someone contested these results!
It’s truly (almost) impossible to conduct these tests across an even playing field. What I could have done was just run one client at a time, with one torrent. But I didn’t want to wait (up to) 15 days for three torrents, (plus, I wouldn’t be able to download my OWN stuff during the tests if I performed it that way). The other reason why this “one-at-a-time” test wouldn’t work is that by the time I got around to the last BitTorrent client, the seed/peers in the torrent would have totally changed, (generally this would lead to an unfair impact on the last client being tested - as the S/L count would be lower, in most cases). Anyways, in these tests nothing else was consuming bandwidth - I had no other torrents, P2P programs or even browsing going on at the same time (as I did these overnight on three separate nights).
I was actually a little surprised by the lack of speed for all results. I have a 1.0MB download connection (8.0 Mbit) and can often get 400-500KB down rate for good torrents (such as these). Only the “BEE MOVIE” torrent cracked the 200KB/s (average) download rate through all clients simultaneously. I forgot to mention that I had Peer Guardian 2 running at the same time, as well. I’ll edit the article and add that in.
This experiment is really more about ‘benchmarking’ them against each other, and see which one leads the pack. To make this test REALLY work, you’d need to have 5 similar computers hooked up to different Internet connections (with each connection being the same “ISP-subscribed” package). Even this would be suspect - no upload/download rates are ever what the ISP tells you, (there are other variables - distance from server; how many “people” are in your area on the same cable ‘pool’, etc.).
Me, personally, I use BitComet for public torrents (mininova, thepiratebay) and I use utorrent for the private sites - but lately I can’t be bothered with BitComet anymore and just use utorrent for all. Often lately I’ve been utilizing ONLY private trackers, due to all the crap on the public sites, nowadays. Most tech-savvy BitTorrent users know better than to use BitComet, except on the public trackers.
—Sharky
03 • Wendy McKee Says: 01.03.08 at 9:51 pm
I can’t thank you enough for this very useful information. I am on dial-up (sucks) but have no choice given that I live in the boonies. Anyway, thanks again. You are awesome for doing all of this work for me (and others who have read).
04 • Trackback :: Bittorrent » Which BitTorrent Client is the Fastest? 04.04.08 at 12:09 pm
[...] sharky wrote an interesting post today on Which BitTorrent Client is the Fastest?Here’s a quick excerptEvery BitTorrent software developer claims that theirs is the fastest BitTorrent client, but which one really IS the fastest? There are so many variables to be considered: Seed/peer ratio of individual torrents, public or private … [...]
05 • Diji1 Says: 20.07.08 at 9:16 am
Another fantastic article on a great site: truly, this is my favourite P2P site handsdown, I’ve learnt so much!
I wonder if testing the clients across a LAN would tell us anything - it would at least eliminate the internet variablity issues.
06 • valavan Says: 05.10.08 at 11:44 am
i think as bitcomet has been included, you should have tried “netexpress”, “flashget” and limewire.
these softwares are on and average 2folds faster than bitcomet.