Increase your file transfer speeds by up to 4x | Hacker News Reader

Increase your file transfer speeds by up to 4x

supertcp.com

10 pointssupertcp29 comments

I assume that since submitter is named "supertcp", you're the author of the site. To you I ask, what exactly does SuperTCP do?

Scrolling down to "How It Works", it doesn't really explain "how it works" but instead, it has vague phrases such as "better algorithms". If one then decides to click on more details via your blog post "Why the Internet is Slow" (which btw often returns "Resource Limit Is Reached"), it still doesn't explain the technical nitty gritty details.[1]

Those vague texts you've written may be ok for Wired Magazine or TechCrunch but for the HN crowd who is more sophisticated about hardcore technology, can you provide a more satisfying description of what it does? Is it jumbo packets? Changing ACKnowledgement timings? What exactly does it do that requires no new software to be installed?

[1]google cache in case the real pages are not responding:

"How it Works": http://webcache.googleusercontent.com/search?q=cache:Jb8UsXV...

"Why the Internet is Slow" http://webcache.googleusercontent.com/search?q=cache:HWoIkC7...

Thanks for the cache links. But this isn't starting well:

"Now, with a direct circuit between you and them the only limit on how quickly you could send data was how you encoded it and the speed of light (or, the speed of electricity anyway)....

"...And apparently somebody else thought of that which is why they built a protocol called the transport control protocol, or TCP...."

A lot of questions are coming in around "nitty gritty details". I'm going to work on a blog post to try to address this. There are some aspects that are proprietary, but hopefully I can answer most–if not all–of these more technical inquiries.

Thanks!

Proprietary? Please note, you won't get a lot of interest from HN for proprietary closed source code.

I finally made it to the blog, and I'm still not at all clear on what you're claiming SuperTCP does.

Is this TCP window scaling? Linux can already do this, not sure about Windows or Mac OS X, though I'd be surprised if they don't.

Is it larger initial window sizes (i.e. WSCALE with larger than normal buffers)? Again, Linux has this RFC implemented, and I'm pretty sure all the other major OSes do, as well. But, I guess this app could be increasing buffer sizes to allow WSCALE to negotiate larger windows.

I have a vague feeling this is snake oil. Possibly a GUI for adjusting some buffer sizes and enabling some options that aren't enabled, by default (because they can cause problems in some environments for some use cases).

If there were something other than hand-waving in the "how it works" and the "technical" blog posts, I might be less suspicious. But, there is nothing but unbacked assertions, seemingly random speed claims, and a pretty slick website.

I don't want to seem overly harsh here...I'm interested in what you're doing, if it is novel. But, extraordinary claims require extraordinary evidence, and you've made truly extraordinary claims. 4x speed up, without compression and without introducing new specifications that have to be implemented on both sides, would be absolutely revolutionary.

So, in the absence of evidence, I am inclined to call, "bullshit!".

Thanks for checking us out SwellJoe!

I think what I've discovered is that the Hacker News community is more tech-savvy than originally thought. Sorry about that! Obviously we're missing some of the technical details of what we're trying to do. I'm going to work on a blog post to help clarify things, but for now you may be interested to know that SuperTCP is essentially a kernel-based TCP proxy. This allows us to–with only changes to one side of the connection–tweak every aspect of the data transmission as long as we stay within TCP's defined restrictions.

Yes, we tweak initial window sizes and lots of other things that can be easily changed on most operating systems, but our main differentiator is how we detect and utilize the available bandwidth, or really how we react to congestion when we see it. SuperTCP tries to be better than the other congestion control algorithms (CUBIC, Reno, NewReno, etc.) most operating systems support. That's where we're better.

I think what I've discovered is that the Hacker News community is more tech-savvy than originally thought.

Understatement of the year.

You try to beat other congestion algorithms, but those are all pretty poor compared to RemyCC. http://web.mit.edu/remy/ How does SuperTCP stand up against it?

RemyCC is new to me but looks fascinating. I'll try to take a look and see how we compare!

Another thing you need to know about Hacker News is that anything like sockpuppetry, astroturfing, and voting for your own posts is not ok here.

For a marketing page, it should really keep the numbers consistent. If SuperTCP can give me a speedup of "up to 4x", then 2.5 hours does not become 15 minutes and 50 minutes does not become 5 minutes. That is a speedup of "up to 10x".

Basic math issues and vague description of the technique make me worry about the technical veracity of the described approach.

I am also worried about fairness: do SuperTCP streams muscle out other streams? Is that how they grab all the bandwidth?

TCP has back-off built-in for a reason: if everyone is as greedy as possible, massive congestion occurs.

Thanks for noticing! We need to clean up that material!

Also, please describe the technical details better---maybe eventually having a whitepaper showing some performance graphs with different workloads?

Criticizing you from a pulpit is easy, kudos to you if you fix things up over time :-)

Yes, we're definitely working on white papers, use case studies and other things that help explain exactly what we're doing to a more technical audience. Thanks!

> Resource Limit Is Reached

Ironic.

I'll bet it was reached in approximately 1/4 time, though.

Is this it?

http://wiki.remobjects.com/wiki/Super_TCP_Channel

No, to be honest I haven't heard of this before. Looks interesting, but not exactly a similar concept. I'll have to look into it.

Its back..

> or upload a YouTube video in 5 minutes instead of 50

I literally don't believe it. Even if you reduce it to pure UDP, you don't save that much from the TCP conversation

Huge spike in traffic. Sorry about that!

I had at a chance to read about your project when the site was still up. Does it require another super-tcp compatible port on the other side (YT, Dropbox, etc...)?

Thanks for taking the time!

No! That's the beauty of it. SuperTCP is one of the only single-sided "better TCP" solutions out there. Put it on one side and you're ready to go.

Sorry guys, site got totally overloaded by a huge serge in traffic. Trying to scale up and get back to everyone....

Since the site is down, I'm assuming fiddling with TCP window sizes is one of the tricks in the bag?

The site is broken on chrome for Android. Still readable, but the alignment is off.

Thanks for noticing! We'll get that sorted.

Flagged. this site tell us nothing about how this works.

Sorry about that! I'm working on getting more detailed technical content out soon to answer all the questions I've been getting.

Reply on news.ycombinator.com