p2pnet.net News:- We recently had an item on Canadian ISP Shaw’s apparent bid to throttle BitTorrent users’ bandwidth during which we also pointed to Rodi, a p2p Java client/host designed to, “serve the filesharing community with fast data delivery and serve the Open Source community by facilitating faster software deployment”.

We had several emails asking (and in one case commanding ; ) us to tell you (if you’re affected) to go here for relief, and a post says the same.

You’ll also find handy hints on other topics.

Read on >>>>>>>>>>>>>>>>>>>>>>>>

To Shaw Fiberlink users

I assume that the Ellacoya promise that the shaper is statefull is true. In simple terms it means that they create an entry for every established TCP connection and keep the record in the database thorugh the whole lifecycle of the connection (i have my doubts about how reliable such thing can be if they do not terminate the connection in their box, but this is separate issue and not related to the current problem). I would post on some popular website the following suggestion to all BT users who suspect that traffic shapers installed on their gateways

Try one of alternatives

• Start download multiple dummy torrents. Configure your BT client to use 1kB/s of upstream and 1kb/s of downstream and up to 2000 connections for every one of these dummy downloads.
• Run SYN port scan using port scan tool like NMAP from http://www.insecure.org/nmap/ for some well known and reliable WEB server like Google or Yahoo.

Explanation. Traffic shaper keeps/records all existing TCP connections or ‘flows’. Because performance of the box is expected to be high they probably use special kind memory like CAM. It is very fast but has limited size. Let check how reliable the box is. I suggest to create multiple dummy TCP connections and run them in the background. You can call it stress test. Every desktop can create about 60K connections simultaneously. i think that reasonable number of connections somewhere on the order of 2-10K. Important ! Every connection costs may be 32-64K RAM depending on OS and TCP/IP stack settings. In case of Windows you will need 512M RAM at least to run such number of connections without signifcant performance degradation. Also make sure that you are not behind NAT – your IP address returned by ipconfig command is the same as one you see here http://www.broadbandreports.com/whois

Average CMTS supports between 20-100K of modems. If 10% of modems establish 2000 conenctions each we are talking about 10M connections. If single record size is 16 bytes we have 160MB data base. There is no way to store it in CAM memory.

If Ellacoya equipment attempts to terminate TCP connection (or PROXY the TCP session) the most painfull for this device is going to be establish of TCP session. My wild guess that they can handle no more than 200K connections/s. It means that for 50K users CMTS we have to establish (and immediately close) 5 TCP connections/s for every one of 50K IP addresses to bring the system down.

Pay attention that depending on CMTS it can bring down the gateway (router) too, but i think that probability is low – packet forwarding is much simpler task than traffic shaping.

From Ellacoya website

The traditional approach to traffic classification on IP networks associates an application with a TCP or UDP port number. For example, HTTP uses port 80 and FTP uses port 21. But this approach fails with peer-to-peer traffic because as much as 60% of peer-to-peer traffic appears on random port numbers. This behavior is commonly known as port-hopping. To accomplish this, however, the peers need to understand what real payload is. For example, if the peer-to-peer desktop application randomly decides to have the traffic appear as HTTP, or web traffic, the other peer needs to understand the incoming traffic is actually peer-to-peer traffic and not web traffic. It’s similar to the way in which encryption systems work – both ends need to understand how to decode the message. The peer-to-peer designers accomplish this by the use of ’signatures’ embedded in the data packet. So, if one knows what the signature looks like and can look inside each data packet traversing the network, one can identify peer-to-peer traffic that are pretending to be something else.

Weak encryption like XOR with key unique for every file (it can be, for example hash of the file) will bring this system to the knees. Such traffic shaper can do absolutely nothing with encrypted data. Actually it can not even recognize that the data is encrypted.

Talking about multiplayer games you guys gave me great idea, thank you.

I understand that you read this website. I think that we could discuss the problem of trafic shaping in the modern networks openly on this website message board. I think that the problem is significantly more complicated than many beleive. It does not mean that there is no solution – there is, but it is not apparent.

From Ellacoya website

In many instances, reducing the amount of bandwidth consumes by peer-to-peer applications can result an additional 25% of the subscribers added to the network without the need for additional CMTS, node splits or other HFC capital investments.

25% is considered to be dramatic impovement ? and what about VoIP, VoD, etc. ? In many applications the traffic is symmetrical, including e-mail. What about personal WEB servers ? In my opinion ISPs have to solve the problem using rational approach.

From Ellacoya website

With the ability to detect and prioritize VoIP applications by subscriber and service provider, the broadband service provider can choose a service delivery model that best suits its needs. For example, once traf.c from a non-facilities based voice service provider is identi.ed, the provider could strike an agreement with the service provider to prioritize its traf.c, or market QoS to the subscriber directly for service enhancement. A large provider rolling out its own VoIP service might guarantee QoS for its service, while leaving all other VoIP traf.c untouched as currently done today.

Does it mean that SBC and Skype will deliver different voice quality ? Do ISPs (broadband service providers) agreements with their customers mention ’service delivery models’ for different service providers. Actually it is very easy to prove that ISP does use traffic shaper. Take any two nodes behind the same default gateway and run different applications on them.

See also dslreports.com Shaw forum.

=============

Something you think we should know? tips[at]p2pnet.net

<--------If at first you don't succeed, get a job at Microsoft——->

See:-
throttle BitTorrent – Is Shaw throttling BT?, p2pnet, April 7, 2005

This entry was posted on Monday, April 11th, 2005 at 5:49 pm and is filed under Uncategorized. You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site.

HOME

Viewed 5721 times