How Big Can an Anonymous Net like ANts or Mute Get?

[Hornet]

During testing ANts users have experienced both fast(50 K b/s) and slow(2 K b/s) download speeds - fast when the network was small(10 users) and slow when the network was bigger(40 to 60 users).

So how big can an ANts (or mute) network become before it is too slow to be useable?

Can these scaling issues be solved or do such networks inherently have to be small?

Jason Rohrer the developer of the original ad hoc ants network application MUTE has tried to answer some of these questions in an interview on http://www.onlamp.com/pub/a/onlamp/2...te.html?page=1:

Jasons says the larger the number of users, lower the average user connection speed and the higher average number of downloads per user then the SLOWER THE AVERAGE DOWNLOAD SPEED OF EACH AND EVERY USER.

If you want uploader/downloader anonymity, you simply cannot use direct downloads. Indirect downloads always involve a substantial performance and scalability hit. Even if you ignore the effect on overall transfer speed, you still have at least one additional node involved in each download, which in turn increases the load induced by each download.

For example, suppose you have a direct download network of 100 nodes that can support 50 simultaneous transfers — half the nodes are uploaders and half are downloaders. If you now force each transfer to involve an additional intermediary node, while keeping similar bandwidth constraints, you can only support 33 simultaneous transfers: one-third uploading, one third-downloading, and the other third relaying.

I will claim that using a single relay or proxy for each transfer doesn't provide enough anonymity. How can you trust your chosen proxy? What if the adversary happens to be operating the proxy that you choose? The same holds true for any system that uses fixed number of proxies for each transfer. If all transfers use two proxies, and you happen to pick two "adversary" nodes to proxy your transfer, your anonymity is compromised.

MUTE uses a variable number of intermediary nodes for each transfer, with the network topology dictating how long each transfer chain is. No matter how many nodes in a transfer chain are controlled by the adversary, the adversary can never be sure that it controls all of the nodes in the chain. Thus, the adversary can never obtain the identity of the uploader or downloader with any degree of certainty.

Since there is no fixed limit to how many nodes a MUTE transfer can pass through, there is also no limit on how much load is induced by a transfer or how slow that transfer will be, and this is where the scalability concerns arise.

Each additional user in the network is likely to initiate additional downloads, which will each increase the load on the network. Of course, if you want decent anonymity, you must make this kind of tradeoff.

To answer questions about how well MUTE will scale, we need to answer other questions first: How slow must a transfer become before it is considered useless? How much bandwidth will the average user dedicate to the MUTE network? How many downloads will each user be requesting? As an extreme example, consider the case in which no one is downloading anything: MUTE can scale limitlessly. At the other extreme, if everyone expects fast transfers and wants to be downloading 100 files simultaneously, MUTE won't scale beyond a handful of users.

Also, I think it depends on how much users value anonymity. A slow anonymous download may be more valuable than a fast download that could land you in court. The same tradeoff operates for quantity: one anonymous download a day may be more valuable than 100 non-anonymous downloads.

As an example, we can assume that a transfer is worthwhile as long as it is coming in at over 5KB/second. If we assume that everyone has a cable modem with a tight upstream bottleneck, then each node can handle relaying or uploading about three files simultaneously. Next, we can assume that each download passes through four intermediary nodes on average. If we have a network of 1,000 nodes, then we can support at most 600 simultaneously downloads at decent rates — each download taxes the upstream bandwidth of five nodes, and each node can handle being taxed by three simultaneous transfers.

Of course, these calculations change as the assumptions change, but we have just laid out assumptions that would suggest that MUTE could support 60% of its users downloading one file each at worthwhile rates. As the percentage of downloaders increase in this network, the download rates would decrease throughout the network.

With the above assumptions, each additional node contributes three transfers worth of bandwidth, but would consume five such units of bandwidth if it were to download. If we reduce the worthwhile transfer rate to 3KB/second, then we achieve a balance: Each additional node can request a download, since it contributes the same amount of bandwidth that its download consumes, so MUTE could support 100% of its users downloading one file each.

If users curb how many simultaneous downloads they request and are content with the resulting transfer rates, then MUTE can scale limitlessly.

However, if everyone else is "playing nice," you can increase your personal gain by initiating additional downloads for yourself and abusing the network. Keeping this kind of greed in check is difficult, especially in an anonymous network.

[notbob]

i bet the 3 people thinking about trying ants just read that and decided against it

[aqlo]

This is a nice honest post though, with actual content.

Shareaza had a big scaling problem back when they were first trying out the G2 network, they seem to have gotten over it (to the extent that have gotten over it) by rigorously testing the supernodes concept until they arrived at something of a happy medium.

To accomplish this though they really did have to have their own private network, when they tried it in the original gnutella network they just pissed everyone off.

I don't know if this is good news though, the supernodes concept seems alien to any attempt at anonymity, for large data trasfers at least.

PS: Well unless you mean the spam font notbob, yeah that might freak out the timid little burrowers.

[Siskabush]

Man, will these fanboys stop spamming thier damn program?

[FreakinWeasel]

Doesn't sound much like fanboy to me Siskabush. The guy is telling you that if there are more than few users on Ants then the network is screwed? Maybe it's just me but if he is selling Ants as a good replacement for any p2p I use, then he needs to work on his pitch!

[MushroomheadXIII]

Originally Posted by Siskabush

Man, will these fanboys stop spamming thier damn program?

It is fanboying, whenever a post is made by them, their program is always included in it, one way or another.

[AussieMatt]

Um this thread was in the Ants forum .A forum to discuss Ants and and technology surrounding it like proxy chaining ,encryption and IP masking and legal issues that arise from using these techniques also technical issses like can the algorithm scale and if the program even works

.Ants is heavilly developed by one developer at the moment and he makes changes and or bugfixes somtimes on a daily basis,I feel sorry for you Win MX refugees that had to wait over a year for any fixes and beta versions from Frontcode.
Having a Ants forum means people are going to disscuss Ants and keeps the Ants threads in one forum. If you dont want to see them on the front page dont respond and they will go away soon enough or complain to a Moderator.Your protests will probably fall on deaf ears seeing the Moderators set Ants and MUTE the forums up in the first place.

[nms04]

not big!! mute is actualy too slow and ant's has connection problems and too few files

[AussieMatt]

MUTE has different routing policys to Ants and no multsource or swarming downloads, so this could be part of MUTE's problem .Jason did this article a few months ago and is reportedly working on MUTE 0.4 so he may have a solution to this problem by now.

EDIT - This Article was written on August 12,2004 Just about when Grwen released Ants Beta 0.0.1

[Hornet]

Originally Posted by AussieMatt

MUTE has different routing policys to Ants and no multsource or swarming downloads, so this could be part of MUTE's problem .Jason did this article a few months ago and is reportedly working on MUTE 0.4 so he may have a solution to this problem by now.

EDIT - This Article was written on August 12,2004 Just about when Grwen released Ants Beta 0.0.1

AussieMatt Jason's article is not about MUTE routing policies. Multisource or swarming downloads just make for a bigger problem.

The issue is not related to MUTE or ANts it is related to proxies. A direct connection uses the bandwidth of one user. A proxy connection uses the bandwidth of several users therefore reducing the capacity of any proxy system. The more hops the greater the reduction in capacity. The bigger the network the more hops the slower the download.

However the real problem is packet loss, due to network congestion caused by bottlenecks in a big network, as lost packets are resent. This makes the congestion worse eventually leading to deadlocks which stop any data being transmitted at any speed!

There are many solutions including:
1)Using network feedback to manage user demand on the network (as well as routing) http://sourceforge.net/forum/message.php?msg_id=2881862
2)Using social pressure to explain to users why they should limit their downloads or not download to enable the network to have spare capacity to proxy. http://www.dsg.cs.tcd.ie/dynamic/upl...tegory54/3.pdf
3)Gwren (ANts creator) is actively working on solutions that regulate node traffic so preventing packet loss and therefore ensuring that bandwidth is not wasted. http://sourceforge.net/projects/antsp2p/

Also, you in the past have suggested clustering of nodes like in Winny to reduce hops and so boost network capacity and speed.

Hornet

[AussieMatt]

Suppose clustering is one reason Winny and Share work so well , and the fact that Japan has a high adoption of broadband so the probelem's releated to bandwidth and bootlenecks due to low upload/download speeds are not a problem anymore .The Share english docs for example state that you shold use a T1 or better to make Share work.No problem in Japan or Korea.

If you want to be a mouthpiece for Gwren you better read what he has to say about scaling issues related to Ants and MUTE first
Grwen on the MUTE forums

I think... this is the reason why swarming, multiple source download, and
rateless systems are indispensable in such nets!

http://sourceforge.net/mailarchive/f...forum_id=37474

EDIT: this is Grwens reponse to the first post in this thread.

[Hornet]

Originally Posted by AussieMatt

Suppose clustering is one reason Winny and Share work so well , and the fact that Japan has a high adoption of broadband so the probelem's releated to bandwidth and bootlenecks due to low upload/download speeds are not a problem anymore .The Share english docs for example state that you shold use a T1 or better to make Share work.No problem in Japan or Korea.

If you want to be a mouthpiece for Gwren you better read what he has to say about scaling issues related to Ants and MUTE first
Grwen on the MUTE forums
http://sourceforge.net/mailarchive/f...forum_id=37474

OK Aussiematt here is my GWREN quote in relation to ANts having both multiple source downloads and swarming:

Originally Posted by GWREN

ANts has both... and uses also a sort of rateless code to optimize
information diffusion,
but I really don"t know if this will be enough with so many adsl connections
around the net :( http://sourceforge.net/mailarchive/m...sg_id=10222653

IMHO the fundemental issue is tha amout of own traffic to network traffic, be that multisource /swarming own traffic or simple download own traffic. If ANts' nodes don't ensure that enough network traffic gets through then the whole net slows then freezes (due to deadlocks).

Note. I think this issue can be solved and that Gwren can solve this issue in a short space of time.

Maybe, until then for fast download speed use ANts in small networks.

Hornet

[nms04]

winny is "anonymous" but fast too, ant's wouldn't be that bad if gwren would rework the connection mechanism ... it would probably be cool if gwren (ants) and jason (mute) would work together so they would be able for sure to create something useful

[AussieMatt]

NMS Antns gives the 3 options to connect not Just IRC you can connect with IRC ,JETI (Jabber Instant Messenger) or Munualyy by adding a trusted peer. Webcaches can be attacked and its alot easier to move to a new IRC server and channel than it is to set up a webcache also you can set up many IRC channels with ants that dont interconnect or bridge back to the Main network with a webcache this would be hard to do.

[ABC_thellookoflove]

slitting the ants network into many ric channels maybe closed is a horrible vision !!
ants is not kdrive ! We need ONE network as well wih hornet, better with mute as well