That said, TCP is not the best we can design given everything we know about designing network protocols. It was good enough for the people that designed it at the time, and possibly (my chronology is fuzzy) was approximately as good as the mathematics would have reasonably allowed when it was developed. We can make it work well enough in many cases -- the economics of inertia. Other narrow use cases are better solved differently but are not general solutions.
It is one of those problems that sounds like it should be easy to solve on the surface but turns into a bloody epic challenge once you start to dig into it. I am not offering a solution, just noting that very few people can.
> possibly (my chronology is fuzzy) was approximately as good as the mathematics would have reasonably allowed
Just because Van Jacobsen's papers spew forth great volumes of mathematics, doesn't mean there is any robust mathematics behind TCP. Read to the end of his paper "Congestion Avoidance and Control". Read past all of the impressive plots graphs and equations. Read to the conclusions.
"The 1-packet increase has less justiﬁcation than the 0.5 decrease. In fact, it’s almost certainly too large."
This statement shows how little formal consideration went into the entire algorithm. The 1-packet increase is not simply too big or too small, it just doesn't make any sense. For starters, how big is the packet? Oh, it isn't defined anywhere. Even if we just go with the de facto internet packet size of 1542 bytes (you know, the old limit for 10Mbit Ethernet)...
Could that one packet increase per roundtrip make equal sense for a 10Gbit path to India with a 400ms round trip time, as it does for a 56Kb link between Berkeley and MIT (his test case)? Of course it doesn't make sense. And it gives lie to any notion that there is a formal underpinning for TCP. They tweaked it until it worked, and then put on a nice mathematical show to feel better about it.
Quoth Van Jacobsen: "We have simulated the above algorithm and it appears to perform well". Oh now I feel better.
Second point, which Braham is covering: TCP makes the assumption that router queue lengths are reasonable. TCP says, fill up the router queues until they drop packets. But router queues have been getting longer and longer as memory gets cheaper. These queues can create additional seconds of delay to layer on top of the 10ms-400ms speed of light delays we see on the internet itself.
EDIT: In that 10Gb to India example, it takes TCP literally DAYS to fill up the pipe because of that "1 packet per roundtrip" window increase. Days, by the way, of no incidental packet loss, because it all gets reset on a loss.
EDIT: I spent 5 years of my life working on the fact that latency was never really factored into the design of most network protocols.
Take a simple example.
The two generals problem proves that, if there exists any nonzero probability of packet loss, two people cannot even coordinate to both have a state 1 at sunrise tomorrow (attack!!!) or both have the state 0 if it is not 100% mathematically guaranteed that they both believe this has been coordinated (since an uncoordinated attack will be a catastrophic loss for them).
(In other words, the guarantee must be such that by sunrise tomorrow state 0-1 or 1-0, in other words one general thinking the attack has been coordinated with certitude and attacks, but the other general thinking the attack has not been confirmed with certitude and does not attack, must be a mathematical impossibility.)
Take a simple approach. The following packets are all encrypted, but any or all may be lost.
1) first general sends: "Let's both be in state 1 tomorrow (coordinated attack). Since an uncoordinated attack is so catastrophic to us, I will only enter state 1 if I receive your reply. Please include the random number 25984357892 in your reply. As soon as I get this the attack is ON. If I don't get such a packet within the hour I will assume this post was intercepted (lost), and I will send another. I will remain in state 0 until I receive that packet."
2) second general sends: "Got your packet with 25984357892. This is my acknowledgment! I will attack as well. In case you don't get this, I know you won't attack thinking I didn't get your message, so I am sending this message continuously."
Great. But what if all messages from the second to the first are intercepted. Now the first thinks all of HIS were intercepted (has received no acks) and doesn't attack, but the second one does. Failure.
So, we have to emend 2) to:
2) second general sends: "Got your packet with 25984357892. This is my acknowledgment! I will attack as well. In case you don't get this, I know you won't attack thinking I didn't get your message, so I am sending this message continuously. In case you don't get any of THESE messages, however, I will not attack. Therefore acknowledge ANY of them with random number 458972984323..."
Ooops. What if all the first general's ack's of the acks are intercepted or lost? (Perhaps the first general is able to send messages until receiving (2), but just as the first general gets 2) conditions change and the general no longer has any of his messages delivered.)
Now the first general thinks he has acknowledged the ack, but the second general doesn't even know if his ack-(cum-request-for-an-ack-back) message was even delivered...
and so it goes...
Of course, in practice you can simply say: "Let's do this for a certain number of acks of acks of acks, 3 let's say, and then just keep sending the same ack to each other, assuming that if the connection was reliable enough to get to three deep, then it will be reliable enough for one of the final acks to make it through." That's a false assumption (mathematically - what guarantee do you have that if 3 of your encrypted messages made it across, at least one of the next 217 that you send by sunrise all with the same message will), but a reasonable one.
So it is not a practical problem. This is a mathematical problem. Although you cannot even do something as simple as "let's agree to both be in state 1 (or neither if we fail to agree), OK?" over a less than guaranteed reliable connection, if the connection has any reliability at all you can get to within a practical level.
once you reliaze that, PROVABLY, you can't even do the most mundane things no matter what, the mathematics the parent is talking about do not seem all that interesting anymore. :)
Uncertainties introduced by packet loss are actually pretty easy to work past.
RED is hard to deploy, so let's change the base protocol instead. - how does that make sense? Everyone would have to start using new libraries and for backward compatibility we'd have to preserve the tcp layer too. That means standards like http would have to get extensions to use SRV records or suffer delays while utp availability is probed.
There's also a complaint that RED will drop packets once the queue is full. I don't get that at all - it will always happen...
In addition I get an impression there is some tension/implied superiority between us (people doing uTP) and them (ones doing RED). Why does it look so ugly? There's a known problem, there's an interesting solution for new software (uTP) and some plan to migrate old protocols transparently (RED). When did that turn into some bizarre conflict and why?
I wasn't complaining about RED dropping packets, just describing how it works.
As for the tension, my point is that my solution works and the other one doesn't. If you want to know why the person I quoted was being such a dismissive jerk, you'll have to ask him.
Is there some reason this was described as a specific behaviour here?
Man you weren't kidding. I tried looking up uTP on wikipedia hoping to come away with some technical understanding. It's full of passive aggressive statements that cite forum posts as their support, with no information on how it actually works. Maybe some of the folks in this thread could go fix that.
The solution is for the end user to intervene, and tell all
their applications to not be such pigs, and use uTP instead
of TCP. Then they’ll have the same transfer rates they
started out with, plus have low latency when browsing the
web and teleconferencing, and not screw up their ISP when
they’re doing bulk data transfers.
With that said, I enjoyed the post. It's an interesting problem, and I do find the base idea attractive: allowing applications to opt to be background traffic.
It's more likely that users will state the problem -- such as, "I want to be able to run WoW and BitTorrent at the same time." From there, the software developers would determine the solution (optimization for latency vs. throughput).
Most of the bandwidth contention is at the edge, right at your DSL connection, so any battle among network connections is a battle among your own usage at any given moment.
Separately, bittorrent is not a latency-sensitive app, it's throughput sensitive, and uTP was designed for bittorent.
The marketing plan is that the because router
vendors are unwilling to say ‘has less memory!’ as a
marketing tactic, maybe they’d be willing to say
‘drops more packets!’ instead. That seems implausible.
The best way to solve that is for a router to notice
when the queue has too much data in it for too long,
and respond by summarily dropping all data in the
queue. /snip/ Of course, I’ve never seen that proposed
We can't get RED or IPv6 deployed, and and the IETF doesn't seem to get anything useful happen these days.
edit: anyone else remember when layer 3 had a bright future ahead of it, IPv6 and end-to-end IPSec (with keys in the DNS) were just around the corner...
But TCP actually does not suck, it's been there for longer than I have and served us pretty darned well up until now.
Never forget that when the TCP protocol was designed, the biggest concern we had was that a nuke would land on top of our heads at any minute and the network should keep working. Also, the "Internet" was thought to be a small niche network of networks among the military and academics.
I guess this is all well known, it's just my reaction to the editorialized title.
Hmmm. Me thinks that TCP does not suck so much.
Anyway, uTP looks cool, LEDBAT sounds very interesting and BitTorrent is of course, completely awesome. I just don't think TCP sucks. I'm constantly surprised at how well it works for how simple most of it is and how complicated and intractable the rest is.
Extremetcp.com is the solution to the congestion problems of TCP. The best part of ExtremeTCP is that it is not a new protocol. It is TCP. It just uses clever algorithms at the sender side to send data while avoiding congestion. (Since TCP does not actually specify which algorithms one should use as long as one avoids congestion, ExtremeTCP is a perfectly legal version of TCP).
Yes, I am involved with this. If interested in testing, please send an email to the contact address in the website.
By the way we do beat the the Linux TCP stack (which now uses Cubic TCP, afaik). The data shown on the site is Compound TCP, which is the current windows implementation.
Let me know if you would like to see the detailed test results.
Unless I'm mistaken, you're not yet at the stage where you're selling a turnkey solutions to pointy haired bosses.
When we say we are the fastest, I mean we are the fastest at sending packets all the way through. This is not easy at all, and it is not as simple as sending packets as fast as possible.
We are very good at modulating our speed in order to have full speed while avoiding dropped packets. Thus, in many situations, we send data slower than other TCP versions but the other versions get into trouble and start dropping packets.
There is one universal rule for TCP congestion avoidance algorithms and that is that as soon as you notice a dropped packet, you have to stop and wait for the congestion to clear up. If you do not do that, you will break the internet. But we do follow that rule; furthermore, we avoid large numbers of dropped packets in the first place.
We have tested our software with other standard connections and it does play well with others.
As someone else noted, Fastsoft are respected by the industry and it is well established that they do not break the internet. We are about 30% faster than fastsoft.
It's trivially easy to modify congestion control to get arbitrarily fast performance in high bandwidth-delay environments. I can tell you from experience that implementing fast performance in extremely lossy environments is harder.
And hardest of all is to come up with a solution that works on a network that is shared with the common congestion control implementations, and that works with billions of end nodes.
I spent years working on this. Feel free to reach out to me on my email address, I'd love to share with you my experience of commercializing such a product.
TCP does suck. If you try to use it for lots of short lived connections. And that pretty much sums up how it's being used nowadays, most the time.
For single, long term connections, TCP is fine.
BitTorrent must not not have any books on copyediting.