BitTorrent Tech Talks are one-hour sessions dedicated to the stuff that keeps us busy / keeps us up at night / keeps us coding. From time to time, we post them here. Because sharing.

In this edition of Tech Talks: an overview of some C++ gems. I threw this talk together because my team was about to start a new project in C++11. Since it’s fairly new, I figured some of it might not be as well-known as it should. Fundamentally, I’m pretty excited about all the new possibilities in C++11. Even higher-level abstractions, at even lower cost than C++98.

In the video below, we go over for-loops, automatic type deduction, lambda functions and more.

Correction: I say that lambdas with an empty capture statement defaults to by-value, which is incorrect. It defaults to not capturing anything.

Follow along with the C++ in the 21st Century slides:


Looking for more on BitTorrent engineering? We’ve got Tech Talks aplenty. Check out Distributed Hash Tables, DHT Bootstrap Update, and Writing High Performance Software.


3.4 is an exciting release for the uTorrent team.

3.4 is the first version to include a major change in the way that uTorrent chooses peers in a swarm. Designed by our own Arvid Norberg, Canonical Peer Priority is a way to help peers connect to the swarm faster, as well as reduce the average hop length from you to any other peer in the swarm.

When a bittorrent client joins a swarm, it needs a way to select which peers it connects to. If it chooses poorly, or if there are malicious actors in the swarm, the connections between clients are not well distributed through the swarm, leading to a large number of hops from node to node. That slows down the ability to each client to pass data on to the next.

You can read a more detailed technical discussion of the issues here, along with graphs and figures that drive home how bad the worst case can be. You can read more about graph connectivity here.

Perhaps one of the biggest changes, though, is one you cannot see. Our engineering team has been growing rapidly, and we have been busy changing our development and release processes. uTorrent 3.4 will mark the first release using improved processes that should allow us to release much more often, while keeping stability at the levels you have come to expect from the world’s fastest and lightest torrent client.

Our previous release cycle was slow. We followed the traditional alpha -> beta -> stable model that a lot of software development follows, for example large video games or operating systems. One of the problems with this style of development is as stabilization work continues on the features you just developed, new features are requested, or requirements change, and now you have to balance two lines of development in the same tree.

Also, with more developers, more changes can be made simultaneously … in theory. In reality, changes in unrelated modules (e.g. the installer) would impact when we could ship new code in other areas (e.g. the disk code), and of course, vice versa. This creates a vicious cycle, where each small problem creates a knock-on effect that impacts other features.

In a situation like this, instead of asking the business to “pick one thing and stick with it” the correct response is for the engineering team to change how they operate.

* On a small scale, picking one thing and sticking with it.
* On a larger scale Multiplexing the work into separate branches.

We needed a way to release changes fast and reliably. This implied quite a few things:
* Don’t mix changes
* Release fast, review results fast

This required us to build a few systems. Some of the larger ones:
* Our release system (code-named “Cherry”)
* Or automatic update system (code-named “The automatic update system”)

It also required programming policies into the smaller parts of the system that already existed
* The build server
* The version control system
* New test servers

These systems, working together, can now answer the question: Is this feature ready for release?
Will deploying this feature likely increase or decrease the crash rate?

We now build individual features in separate branches, which are automatically tested for stability before being integrated into the mainline. That gives us confidence that we won’t slow other engineers down, and that we won’t release a low-quality build to customers.

This effort would not have been possible without the support of the excellent engineering team at Bittorrent.

I look forward to covering these in detail in later posts.

From the uTorrent engineering team, and the rest of Bittorrent as a whole, Happy torrenting!

BitTorrent Tech Talks are one-hour sessions dedicated to the stuff that keeps us busy / keeps us up at night / keeps us coding. From time to time, we post them here. Because sharing.

In this edition of Tech Talks: BitTorrent’s Chief Architect Arvid Norberg describes how modern computers work, and the key challenges to software performance. He then breaks down how to modify data structures to better take advantage of your memory cache. Want to write your own high performance software? Watch the full video, or follow along with the slides below.

Writing High-Performance Software by Arvid Norberg from bittorrentinc

For more talks by Arvid, check out distributed hashtables and his follow-up, distributed hashtables updated.

Update on BitTorrent Chat

Abe —  December 19, 2013 — 59 Comments

Inside BitTorrent’s approach to building serverless messaging apps.


First, a few words on Chat’s origins. Here at BitTorrent, we value privacy. With the news this year reminding us all of the susceptibility of the communications platforms we rely on to snooping, we found ourselves wanting something new, something secure, something private. We ultimately realized that we were uniquely qualified to build this platform.

The primary weakness that we see in the available communications platforms is that they all rely on some central server to route and store all of your communication. Even if your provider can deliver industry-standard security, they cannot provide you with any kind of assurance that your communication is private. All it takes is the right (or wrong) person gaining access to your provider’s central servers, and your privacy evaporates.

Enter BitTorrent Chat. We’re building a product that allows you to talk to your friends using peer-to-peer. No central authority required.

Continue Reading…

DHT Bootstrap Update

arvid —  December 19, 2013 — 2 Comments

Arvid Norberg, chief architect for BitTorrent, Inc, introduces a new DHT bootstrap server. This latest version introduces Node ID enforcement as an important step in our development for BitTorrent Chat. It’s also now open source so that anyone can run their own bootstrap node.

The BitTorrent Distributed Hash Table (DHT) has a fundamental dependency on being introduced to some nodes that are already in the network. There are many sources of these nodes. For instance, your client is likely to save nodes on disk to retry them when you start back up again. Any BitTorrent peers are likely to be on the DHT as well, so those are also tried. However, if you just installed a BitTorrent client, and you don’t have any BitTorrent peers, you must rely on a bootstrap server.

BitTorrent Inc. runs ```` on port 8991 for this purpose.

We are now providing our DHT bootstrap server open source on github. You can now run your own DHT bootstrap node! Please play with it and contribute fixes, features, and performance improvements.

The DHT bootstrapper has some interesting properties. Up until 5 years ago or so, ```` was running just another DHT node, just like the one in µTorrent. This had some obvious problems. Since the default routing table size is 8 nodes per bucket, half of all requests to the bootstrap would get the same 8 nodes handed back to it. At several thousand requests per second, this would effectively DDoS any poor node that happened to end up in its routing table.

We rewrote the bootstrap server to have a flat array of nodes instead and to have two cursors, one for reading and one for writing new nodes into it. Every node that pings the bootstrap server is put in a queue and pulled out 15 minutes later to be pinged. If it is still alive, it is added to the node list.

This is still the case with the latest rewrite, with one addition: Node ID enforcement. We have been looking at securing the DHT, making it harder to attack (especially with sybils). One thing we’re implementing to support this is requiring DHT nodes to calculate their node ID based on their external IP, with some flexibility to support NATs and such. More info on Node ID enforcement can be found here.

The idea is that with Node ID enforcement sybil attacks, where one machine pretends to be thousands of nodes, will become impossible.

The new bootstrap server will still serve nodes with invalid node IDs (in fact, legitimate nodes just joining are not likely to know their external IP yet). However, it will not ping nor add these nodes to the node list for handing out.

This is one step in the preparations we’re making for BitTorrent Chat, which will rely on the DHT and benefits from having a DHT that’s harder to eavesdrop and scrape.

An Nguyen, developer at BitTorrent

An Nguyen, developer at BitTorrent.
(An wants you to know he wears a tank top most days, and not just for blog photo shoots.)

In Tech Talk, we share the stuff that keeps us busy, keeps us up at night, and keeps us coding.

Often, we have a bunch of data that we want to view or edit over and over again. Here’s how to get started with a graphical user interface (GUI) tool to simplify your workflow, help you work more efficiently, and prevent errors along the way.


If you’re on Windows, native on a PC or via VM/virtual machine, you can leverage Windows Presentation Foundation (WPF), a free Microsoft GUI framework with many widgets and controls out of the box. Others are freely available here.

Continue Reading…

uT Linux Server
Good news!

We’ve just updated the µTorrent Server for Linux. It’s a pretty major refresh from the last version. What’s new? Here’s the rundown.

New features:

Touch based input option for Web UI: use your tablet or mobile phone to control your torrent client in the home network integration: use your tablet, mobile phone or computer to control your torrent client from anywhere on the Internet. Either visit or use the Android µTorrent Remote application.

Big improvements:

Performance enhancements to the Torrent engine: the faster just got faster.
Improved web UI: a modern rewrite of the UI for controlling the torrent client in the home network
□ Support for signed torrents
□ Updated RSS feed parser
□ Multiple bug fixes and enhancements

Give the new version a shot, and let us know what you think. Our forums are open. All questions, ideas, and feedback are welcome.


–Team BitTorrent

Screen Shot 2013-04-23 at 8.06.24 PM

Brave inventors, intrepid tinkerers, awesome doers, and startups everywhere: we’ve got a job for you.

Over the course of the past few months, we’ve been busy in the Lab. We’ve been hard at work on a solution for personal file management; opening up an Alpha version of BitTorrent Sync. We’ve been looking at ways to streamline creative collaboration; using a simple file delivery service we call SoShare. We’ve been exploring search prioritization; helping artists find visibility via BitTorrent Surf. And we’ve been experimenting with the storytelling and monetization potential of BitTorrent publishing; via content releases with media innovators: authors like Tim Ferriss, studios like Cinedigm.

Our goal is to create a more sustainable distribution model for the Internet’s creators and fans.

We’re in this together.

And we want your help.

We’re opening up our innovation lab to new media startups. If you have a publicly available product in sound, film, live streaming, media sharing, or peer-to-peer technologies, we’ll work with you to accelerate your idea’s success. Continue Reading…

There are a couple of fairly nasty bugs in the LSP sample code from Microsoft in the function WSPSelect, and as far as I can tell, this one hasn’t been reported anywhere before. Since we’re working on an LSP here, I’ve had to fix this bug. :-)

The bug affected iTunes and safari, which I have noticed some traffic about.

The actual bug is in the three clauses that copy the activations from the provider-socket fd_sets into the client-supplied fd_sets that contain client-visible socket objects

They look like this:
(note that ReadFds is internal, and contains provider sockets, readfds is the user’s parameter, and contains client facing socket descriptors)

And there are three copies of this code for the three fd_sets passed into select.

So what happens if we pass two fd_sets to this code with the same address? Note that in each clause (seperately), the client’s fd_set is zeroed!

So WSASelect(0, &my_sockets, &my_sockets, &my_sockets, NULL) will report the correct count, but only sockets that have exceptions will remain activated when we return to the caller! Moving the FD_ZERO macro expansions and the count captures all above these passes through the fd_sets will fix this bug.

The second one is less severe but still might be harmful; note that the return from the lower provider’s select is a count of activated sockets. This count is not the number of fd_set members that achieved activation, but the number of sockets, so if a socket is repeated for some reason in the incoming fd_set, it will only be counted once on input, but it will be represented twice in our internal array. Microsoft’s sample code will terminate early in this case (note that HandleCount– will be executed twice for one socket), with two activations of one socket, but leaving any remaining events uncopied, My suggestion is to also check whether the we’d be setting the same socket twice into the client’s fd_set, so we can skip double counting as the underlying provider apparently does.

In any case, you should definitely check the code you’re importing. Even OS authors make mistakes, as much as their frequent machismo might suggest otherwise.


Now that we have begun to send out invites to BitTorrent Live broadcaster applicants, it seemed like a great time to begin a series of technical articles discussing the product. We’re taking slow steps toward completely opening the floodgates, trying to make sure that we have a usable, rock solid product once we release it to the world. Right now, you can enjoy the channels of our first broadcasters without an invite.

Meanwhile, expect to see explanations of how Live operates internally, as well as tutorials on how to put it to use, here on the BitTorrent Engineering Blog.

The development of BitTorrent Live has consisted of work in essentially two camps. First is the Python-based ‘core’, which handles the actual functioning of the protocol, as well as the underlying cryptography used when transmitting and receiving data. Second is the ‘app’, which consists of a tight layer around the core, providing a command line interface, as well as a layer beyond that which provides a Windows, Mac, or Linux GUI.

As the core has become more stable over time, and major changes to the protocol less frequent, providing features like Local Peer Discovery has become a possibility. It is present in all major BitTorrent clients, and provides great performance and efficiency gains. Live is an entirely new protocol, written from the ground up, so adding support required a bit of planning.
Continue Reading…