Posted on 03/24/2008 7:59:53 AM PDT by TChris
Bob Briscoe (Chief researcher at the BT Network Research Centre) is on a mission to tackle one of the biggest problems facing the Internet. He wants the world to know that TCP (Transmission Control Protocol) congestion control is fundamentally broken and he has a proposal for the IETF to fix the root cause of the problem.
The Internet faced its first congestion crisis in 1986 when too much network traffic caused a series of Internet meltdowns when everything slowed to a crawl. Today’s problem is more subtle and lesser known since the network still appears to be working correctly and fairly. But underneath that facade and illusion of fairness, a very small percentage of users hog most of the Internet’s capacity suffocating all other users and applications.
(Excerpt) Read more at blogs.zdnet.com ...
If memory serves, that was due to the Morris worm.
Personally, I think a lot of net.congestion could be eliminated if we simply lopped off China from the 'net. All they do is use it to attack our defense and infrastructure networks anyway. Screw 'em.
There's plenty of P2P traffic on the 'net from the US and Europe. I don't think this particular issue is China's fault.
Read the article, it's a good one.
/mark
Please.....P2P isn’t half the problem that rampant spammers are.
How do sites like YouTube stream video? Are they opening up multiple TCP ports to the client like a P2P connection?
What you say is true. P2P has become something of a scourge, but compared to the botnets, it’s not much more than an episodic annoyance. That said, on my mention of China, you should see my firewall logs. I have more bits twitching on the floor from China than any other nation’s netblocks. Even Romania hardly counts as a blip on the radar compared to the Red Chinese.
I would expect that commercial/server connections would be treated a bit differently. Servers need to have access to, and pay for, massive bandwidth 24/7. The rank-and-file home user does not.
Put a price on the users' total bandwidth usage (and allow them an easy way to track it) and suddenly people won't find it so desireable to download and host hundreds of movies and songs.
I haven't read anything that analyzes the bandwidth impact of spam. I wouldn't think that it's all that bad, overall.
All email combined isn't a large percentage of the Internet's traffic. So even if spam is 80% (wild-guess percentage) of all email traffic, it still wouldn't have a huge impact.
P2P is a major bandwidth hog, and it's a long-term one, with hours and days devoted to huge media transfers. EMail is delivered in relatively short bursts.
I agree with you, but I don't think those desires are a bad thing.
A company should be free to restrict its own services that may weaken demand from more profitable business. Customers are free to change ISPs or drop their service if they so choose.
If another ISP wants to invest in the infrastructure to compete with the telcos and cable companies, offering completely unrestricted access, it is certainly free to do so.
TCP in not unfair. Every user of TCP/IP has the same tool available to them. Every ISP has access to the same equipment and the same bandwidth restriction or queuing techniques.
ISPs use different types and levels of queuing to create the congestion environment and base traffic load in the manner that best fits their desired customer base.
Other technical elements that have not been discussed:
1) An end user is limited by their connection speed. User A above, if connected by ISDN, is still limited to ~120K of bandwidth REGARDLESS of how many streams are opened.
2) ISPs can and sometimes do, throttle bandwidth to match contracts. As part of these contracts a “burst” of traffic is allowed. However, these bursts are regulated and when the burst becomes sustained, the tokens that allow for bursting run out and the user is throttled back to their original speed. This is the MOST COMMON reason P2P are throttled. It is also the largest cause for complaints.
3) P2P networks can use other protocols other than TCP. IPSec tunnels, UDP, and other protocols are all possible. So “fixing” TCP will not necessarily fix the problem as P2P software will simply migrate to other protocols but still take up the same, OR MORE, bandwidth
4) Most congestion control techniques (throttling) do not even kick in until a congestion event (router interface overrun) occurs. This so rarely happens in the core of a carrier's network that most cores do not even implement congestion control. The most likely place for congestion to occur is at the edge of the network - that "last mile" connection device.
5) Dropping packets is a natural part of a burst of too much traffic hitting an interface at a moment in time. So much so that there is a lag on the router because it takes a little longer to decide where and then forward the traffic to the correct interface. To help reduce this, carriers implement buffers that provide a small pool of packets that can be handled before the router has to start dropping packets. However, in a congestion event, routers have the ability to selectively remove "drop" packets out of their buffers. This helps by allowing the router to drop packets of less importance and thus maintain service to either more important traffic or across a wider set of users.
At issue is ... what is important or fair? If you just use random packet drop in the example the article presents, then the P2P users is many times more likely to suffer a packet drop on one of their streams and thus throttle their traffic. If you drop the P2P traffic first, or more frequently, then other users, who are not using P2P, will continue their traffic and their packets will receive "full service". Once their packets get out of the way, then the P2P packets will resume.
This is a POLICY / COMMERCIAL decision. There are many other means of either dropping packets or restricting bandwidth. Focusing on one protocol to address one tiny issue that is a POLICY issue, not a technical one is a fool's errand.
Spam, on the other hand, is amazingly high….stunning….just my domain rejects an average of 120,000 to 160,000 spams every 24 hours, and I only have 170 mailboxes. On average, I’m rejecting 104 spams a minute…more than one a second. THAT’S overhead.
Sure, the spams are small but they’re relentless, and require more internet resources to process than a steady P2P stream. Spam is the biggest problem on the internet right now, not P2P.
In general I agree with you, though in this particular case consumers don't always have multiple companies to choose from.
A simple solution for the ISPs is to stop selling "unlimited usage" to customers when you have no intention of providing it. In other arenas that's called "fraud".
Interesting read, but this takes control out of TCP and puts it at the application level. Which strikes me as begging for abuse. Who thinks the same people who realized the obvious way around AIMD, aren't going to realize the obvious way around this?
It's not just email spam that's the problem for me.
About half my web site traffic is either attempts to post spam in a blog/message system or referrer spam trying to get a referral URL listed.
There's no posting without registration so that fails and I don't post referral info so they're not getting anything out of that either.
But it doesn't matter to them, they're using zombie machines, they just keep knocking on the door eating about half of my total bandwidth.
It's my biggest problem. Dependable bandwidth isn't cheap to to pay for and accommodating spam bots frys my bacon pretty good.
Disclaimer: Opinions posted on Free Republic are those of the individual posters and do not necessarily represent the opinion of Free Republic or its management. All materials posted herein are protected by copyright law and the exemption for fair use of copyrighted works.