0MQ

0MQ is a lightweight messaging kernel.

Getting two apps to talk is always a challenge. They have to agree on what the data looks like. And what it means. They have to agree on how the data is organized into a message. XML, binary, or something else? They have to agree how to speak to each other. They have to agree on security. They have to agree how to connect, across what networks. They need to queue messages that can't be delivered immediately. They need to detect and handle errors properly.

Many apps try to solve these problems from scratch. But you can also buy big, expensive, slow packages called "messaging systems" that handle a lot of these jobs for you. Some of these systems even agree on standard protocols like AMQP. Using a messaging system usually costs a lot but it gives you the power to bring apps together.

Messaging used to be for big wealthy firms only, and the really fast messaging systems are still very expensive. This situation is changing.

Here are some key features of "0MQ":

· Our mission with �MQ ("ZeroMQ") is to build the fastest messaging ever. The way to get performance is to optimise the whole software and hardware stack together. So, we are developing �MQ closely with major hardware firms.
· �MQ is already very fast. We're getting 25μsec end-to-end latencies and 2.8 million messages per second today.
· �MQ is completely free GPL-licensed software written in C++.
· �MQ is a project in progress. Download the code, read the whitepaper and test results, or join the project if you are a messaging fanatic.

�MQ is a collaboration between a group of firms, namely FastMQ (which brings experience of the financial messaging market), iMatix (which brings the AMQP expertise), Intel (which brings hardware expertise) and other firms with hardware, operating system, network, and application expertise. FastMQ Inc. will unify all copyrights on the code.

�MQ Goals
· Built on standards: for frameworks, protocols, and community.
· Demolishes any network, from 10Gb Ethernet, to 4X IB.
· Transfers 2M messages a second on a 1Gb network with one CPU at each end.
· End-to-end latency under twenty μsec over TCP using custom network hardware and drivers.
· Beats any commercial product: faster, more portable, more reliable, more usable.
· Built for the community: free software in an open source community.
· Designed for the real world: video, stocks, gaming, grids, real time data, streams.
· Runs everywhere: the smallest devices, all operating systems, all languages.
· Fully distributed: no central servers to crash, billions of WAN and LAN nodes.
· Extensible framework: kernel-style drivers for custom hardware or applications.

What is new in this release:

• Bug fixes:
• Fixed issue 223, assertion failure in tcp_connecter.cpp:300 when connecting to a server that is on an unreachable network (errno is equal to ENETUNREACH).
• Changes:
• Removed Debian packaging, which is now available at packages.debian.org or via apt-get.

What is new in version 2.1.3:

• Fix to PUSH sockets, which would sometimes deliver tail frames of a multipart message to new subscribers (Martin Sustrik).
• Fix to PUB sockets, which would sometimes deliver tail frames of a multipart message to new subscribers (Martin Sustrik).
• Windows build was broken due to EPROTONOSUPPORT not being defined. This has now been fixed (Martin Sustrik).
• Various fixes to make OpenVMS port work (Brett Cameron).
• Corrected Reference Manual to note that ZMQ_LINGER socket option may be set at any time, not just before connecting/binding (Pieter Hintjens).
• Fix to C++ binding to properly close sockets (Guido Goldstein).
• Removed obsolete assert from pgm_socket.cpp (Martin Sustrik).
• Removed stand-alone devices (/devices subdirectory) from distribution. These undocumented programs remain available in older packages (Pieter Hintjens).
• OpenPGM default rate raised to 40mbps by default (Steven McCoy).
• ZMQ_DEALER and ZMQ_ROUTER macros provided to ease upgrade to 0MQ/3.0. These are scheduled to replace ZMQ_XREQ and ZMQ_XREP (Pieter Hintjens).
• Added man page for zmq_device(3) which was hereto undocumented (Pieter Hintjens).
• Removed zmq_queue(3), zmq_forwarder(3), zmq_streamer(3) man pages (Pieter Hintjens).
• Upgraded OpenPGM to version 5.1.114 (Steven McCoy, Mikko Koppanen).
• Build system now calls OpenPGM build process directly, allowing easier future upgrades of OpenPGM (Mikko Koppanen).
• Build system allows configuration with arbitrary versions of OpenPGM (./configure --with-pgm=libpgm-x.y.z) (Mikko Koppanen).
• OpenPGM uses new PGM_ODATA_MAX_RTE controlling original data instead of PGM_TXW_MAX_RTE covering entire channel (Steven McCoy).
• 0MQ builds properly on FreeBSD (Mikko Koppanen).

What is new in version 2.1.2 RC2:

• 0MQ now correctly handles durable inproc sockets; previously it ignored explicit identities on inproc sockets.
• Various memory leaks were fixed.
• OpenPGM sender/receiver creation fixed.

What is new in version 2.1.1:

• This a stable release with several important bug fixes over 2.1.0 and no major functionality changes.

What is new in version 2.0.10:

• This is a bugfix release with no major changes.

What is new in version 2.0.9:

• Fixed several bugs.

What is new in version 2.0.8:

• The ZMQ_SWAP functionality has now been implemented for all socket types.
• About 30 bugs were fixed.

What is new in version 2.0 Beta 6:

• Add Git location to README
• Update README
• Add MSVC build files to distribution
• Removing leftover MSVC builds
• Merge branch 'master' of git@github.com:sustrik/zeromq2
• Clarify zmq_poll restrictions
• C++ interface documentation updates
• Cleanups to autogen.sh
• Use POSIX "command -v" construct to test for prerequisite commands
• Clarify error messages on failure
• pipe_t::rollback removes only unfinished message from the pipe rather than all unflushed messages
• Implement flow control
• This commit introduces the necessary changes necessary
• for implementing flow control. None of the socket types
• implements the flow control yet. The code will crash when
• the flow control is enabled and the thw lwm is reached.
• The following commits will add flow-control support for
• individual socket types.
• C and C++ headers moved from bindings/ to include/, bindings/ removed
• --with-c and --with-cpp options to configure removed, C and C++ now built
• and installed by default

What is new in version 2.0 Beta 1:

• This version includes Common Lisp API, zero-copy for large messages and more.