windows vs. linux

[pondini]

a friend posted the below text at another forum. i'm curious to know the specifics behind the fluctuations on each OS.

I took advantage of some "forced" vacations to try Linux for torrenting.

I've been messing around with Azureus and uTorrent to try to find which is best... And now I know... Under Windows everything sucks

I set up Ubuntu on a VM under VMWare, "et voilá".... I finally got the bandwidth.

From the same seeds I get 5 to 10x the speed I got under windows when leeching. (yes... 40KBps under uTorrent.... 360KBps under Ktorrent)

When seeding I got "ALL" my BW to be used on a single torrent.

[jewelisheaven]

It has very little to do with the OS and more to do with settings.

On the exact same torrent speeds will be the same, unless peers do not communicate (i.e. within the disparate DHT swarms)

Edit: I don't even know if this is an issue anymore? I regularly see Azureus and mainline clients with "H" flag (gathered through DHT)

[Firon]

You do realize that Comcast is a horrible ISP that blocks BT and various other things, right?

[jewelisheaven]

Noone mentioned comcrap (in this thread) boyo.

And yes, I did. It's marginally better on TimeWarner's Roadrunner if only because they flat-out tell you you don't get unlimited bandwidth and are restricted during certain hours.

[Firon]

That's his ISP.

[jewelisheaven]

I forgot you were psychic Firon, apologies.

Edited because I must have been sleep-dep

[pondini]

yeah firon, i was wondering if it was a traffic-shaping issue as well. i posted the below text in the other thread before i posted here...

xxxx, very creative experiments you have been conducting:) good job:) and i tend to agree that it is probably network management, altho i'm curious if ktorrent is NOT identified as a p2p app [packet signatures] and escapes traffic shaping. i'm probably just paranoid, but i'd love to hear what they have to say about your findings at the utorrent boards;)

[matamoscas]

I'm the one "experimenting"...

I can't say that it's not my ISP.... All I can say is that running torrents under linux, beats running them under windows by far and large... at least in my ISP.

I don't know if screenshots to ilustrate the point are welcome here, but if they are, I'll be happy to post them.

I can add that Linux was Ubunto 10.7 distro, out of the box, running on VMWare.

[Firon]

Nah, it's almost certainly your ISP. Deluge does something different to avoid traffic shaping as well (while still keeping the same encryption method), but we still haven't figured out what exactly.

[Switeck]

What's Deluge's half open retry rate?

Are you firewalled in Deluge?

[matamoscas]

I was connecting from an ISP in the EU, the torrent was the single torrent running on the system with three seeds all under Utorrent and all in the US. (nothing else hogging bw on systems)

1 seed probably was capped (8/10KBps) and was the same althrough the tests.

other seed was probably using all seeders available bw and running at around 80KBps in all tests

Third seed run at about 85KBps under XP/uTorrent and at about 240 KBps under Ubuntu/Ktorrent.

This was the second batch of tests... the first, was run against a single seed in an EU server (2k3/uTorrent) and yelded even greater differences.

If I get to have time to test it on other ISP I'll let you know.

[Firon]

You did use identical settings, right?

And.. you weren't running both at the same time, right?

[matamoscas]

I don't know what settings you refer to. (I have XP/sp2 connection limit patched to 200 now and uTorrent set to a max of 50) I'm not a Linux guy (haven't used it since college days... a long, long time ago) so I made no changes whatsoever there.

I'm not capping BW either up or down in either setup nor limiting number of connections or upload slots in bt software. Using a port on 5995x range.

No, I wasn't running both at the same time... stop 1, start the other...& vice-versa. And always single torrent active, and no other running apps.

If you would like to perform some tests yourself to either setup, that can be easily arranjed. Please feel free to e-mail me about it.

I made a few more tests today (now using CentOS 5 and Ktorrent 2.2 on VMWare 6.2) and it continues to yeld great differences.

[Firon]

You need to configure upload limits, connection limits and upload slot limits for µTorrent.

[matamoscas]

Currently I have:

Global conn: 200

Max peers per torrent: 200

Upload slots per torrent 25

Besides that, I changed two parameters in the advanced tab:

I have set bt.allow_same_ip to true and net.max_halfopen to 50

I have a connection that is effectively 10Mbps downstream 1Mbps upstream.

The system is a 2.4 Ghz PIV with 1GB.

What values would you sugest?

[Switeck]

You probably don't need net.max_halfopen to be 50 (trying 50 new ips at once!) if you're only going to allow 200 total connections. This is especially true if you're not firewalled and have no problem receiving incoming connections in µTorrent!

If you're running more than 1 torrent at a time, 25 upload slots per torrent is probably a bit too high too...Speed Guide (CTRL+G) only recommends 6 upload slots across 6 total torrents (for 36 total upload slots at 92 KiloBYTES/sec upload speed for 2.55 KiloBYTES/sec average upload speed per upload slot). 25 upload slots is ok though if you only allow 1 torrent at a time.

My recommendations:

Global connections: 200

Max connections per torrent: 80

Upload slots per torrent: 6

net.max_halfopen: 8

upload speed max: 90 KiloBYTES/sec

alternate upload speed max: 110 KiloBYTES/sec (may be a bit much...so do test!)

total torrents at once: 5

max downloading torrents: 3

[matamoscas]

Tnx for the recommendations Switek.

I'm going to try those settings... I'll omit the upload limitations unless you advise otherwise, since I seldom download and while testing I have no other networking app running that could be impaired. So unless having them set to 0 impairs uTorrent performance, I'll leave it as is.

Edit:

I tried with the modified setting, but couldn't see any improvement.

I noticed the linux client sets IP TypeOfService to 8(dec.)

[Switeck]

Yes, having upload speed set to 0 impairs µTorrent's performance.

[matamoscas]

That seems to have made some difference yes. Tnx.

There was a single torrent beeing leeched at about 12KBps and is now at about 18KBps

Later on I'll make some more tests.

Edited:

I've made the "some more tests" and uTorrent with the correct settings has had about 60% increase in performance.

About the great speed diferences previously pointed I found this:

Checking on both the seeder and leecher side, I found that kTorrent 2.2 is very "optimistic" on the peer connection speeds, and values presented on the client were much larger than the ones I could see on server side...

So I changed the test a bit and started evaluating how much data was transfered during a 300 sec. interval.

Once again single active torrent and single active app on the system at the time.

first run:

Windows/uTorrent 98MB (260KBps)

Linux/kTorrrent 125MB (416Kbps)

Second run:

Windows/uTorrent 104MB (346KBps)

Linux/kTorrrent 146MB (486Kbps)

Third run:

Windows/uTorrent 113MB (376KBps)

Linux/kTorrrent 163MB (542Kbps)

So the difference is not so big... but Linux still performs better.

[uAzu]

perhaps the difference is your linux OS is clean whereas your windoze OS is cluttered with malware, registry bloat, and pointless network "tweaks"?

Compare your current state with a NEW install of windoze

[jewelisheaven]

sr. butterfly please refrain from inflamatory words

uTorrent has its perks, for example low memory footprint, cross-platform usability with the SAME EXE which itself is under .25 MiB. Also stress testing it it can handle over 2000 currently loaded/running torrents with only ~ 140 MB usage. Try to say that about Azureus.

Also if you want to tout Ktorrent, it doesn't support encryption, at least all of the various versions I have connected to in my years running uT. Though it is essentially becoming a dying setting which doesn't increase throughput, there are still places out there which benefit from such a feature. Perhaps you can ask the developers to implement it.

In any case, people will use what they prefer, or in the extreme, when something makes a LARGE impression upon them they may switch (people are innate at brand names)... so to each his own

[matamoscas]

Might have some... can never tell for sure... But win OS install is about a month old, updated, and running firewall, Mcafee ans SpySweeper... About "pointless" network tweaks I have two:

TCPIP.SYS patched to allow more than the std SP2 max. of 10 half-opened connections, and the MTU set to PPOE size.

If you can advise otherwise, I'll rerun tests.

I make absolutelly no question of running Linux.. as it happens its behaving better for bt.

But just to make sure and make tests absolutely even, I'll also setup a XP VM and do some more testing. (it will take a few days though since I no longer have much time at hand.)

[jewelisheaven]

understand even with equivalent settings virtualization still involves overhead so it is not a true test for equality.

But before didn't you see The main point I wanted to get across is "to each his own" meaning people will use what they want to use, no matter what information to the contrary.

[Switeck]

You never mentioned if your upload speeds were stable in µTorrent.

BitTorrent lives, breathes and DIES by how well it can upload!

...so this is kinda important.

Lower half open rates for µTorrent (due to windows limits) will result in at least a slightly slower ramp-up to max speed than probably possible in ktorrent. However when you are downloading "typical" torrents that have fewer than 50 total peers+seeds...and are over 1 GB in size...then a lower half open rate can result in slightly higher download and upload speeds. And it's not as hard on marginal networking software and hardware.

[matamoscas]

Global upload speed is steady...Peer upload speeds may have significant variations.

But I can only reach the max set speed if I have several leechers... I don't recall a single leecher ever filling available upstream bw.

Thats what got me so excited with Linux in the first place. It was the first time I saw a single connection using all bw... (after what I found out about reported speeds on Ktorrent 2.2, I'm no longer so sure it was true....) but thats a test I'm going to repeat.

But as I said before I'll do it from a XP VM I still have to setup. So it may take some time before I give some feedback.

The tests I made were all with few connections... In the last one the torrent had 6 seeds and no other leechers, most of them were with torrents with a single seed, seeding from a 100Mbps source.

Tnx for beeing so cooperative. I've already learned some things. Hope we can profit even more.