Spines
Version 5.5 - December 23, 2020
- The main change is this version is to use OpenSSL version 1.1.1
- Allows use of key size 512 bits in addition to previous version's 1024 bits and 2048 bits
Version 5.4 - November 26, 2018
- The main change is this version is to the dissemination-graph-based routing algorithm
- Support for dissemination-graph-based routing in Spines 5.3 used
static pre-computed graphs in all cases, but switched between these
graphs based on changing network conditions. The 5.4 release makes
this approach more dynamic, using two dynamically computed (not
static) node-disjoint paths in the normal case, and including dynamic
node-disjoint paths as one component of the resilient graphs used
under the most problematic conditions.
- This approach is described in Amy Babay's 2018 PhD thesis, "Timely,
Reliable, and Cost-Effective Internet Transport Service using
Structured Overlay Networks"
- Fixed bug in the interaction between link encryption and rate limiting
added in Spines 5.3 that could cause crashes
- Reported by Tiger Gao, Daniel Qian, Elaine Wong, and Jason Wong
- Fixed bug in realtime link protocol's calculation of how many nacks can
fit in a packet
- Additional minor bugfixes, improved enforcement of message size
constraints
- Performance tuning, increasing maximum message sizes and default
bandwidth allowance (when rate limiting is used)
Version 5.3 - March 9, 2018
- Added Source-Based Routing capability (separate from K-paths used in
Intrusion-Tolerant Messaging)
- Supports using K node-disjoint paths, flooding, or dissemination graphs (see below)
- Added support for dissemination-graph-based routing (a specific type of source-based routing)
- Based on the protocol from: "Timely, Reliable, and Cost-Effective
Internet Transport Service using Dissemination Graphs," Amy Babay,
Emily Wagner, Michael Dinitz, and Yair Amir, in Proceedings of the
IEEE 37th International Conference on Distributed Computing Systems,
Atlanta, GA, June 2017, pp. 1-12.
- Turned off TCP fairness by default (can be enabled by starting Spines with
the -tf commandline flag)
- By default, we now turn off TCP fairness but impose a 500 Mbps rate
limit on each link. The rate limit can be changed or turned off using
the commandline -rl option
- Added an option for tracking and periodically printing the number of
messages sent for each destination client at each daemon
- Simplified loss rate calculation
- Changed report format of spines2port and port2spines test programs
- Fixed memory leak in intrusion-tolerant protocols stemming from not
cleaning up OpenSSL context structures correctly
- Fixed libpread-util memory system to support allocating types smaller than
sizeof(void*)
- Bug fixes in the reliable session
- Some tuning of default timeouts and window sizes to support higher
throughput and more aggressive rerouting
Version 5.2 - May 17, 2017
- Added configuration option to encrypt messages sent on intrusion-tolerant data link
- Added configuration option to enable or disable remote client connections to daemons
- Hardened code (security-wise) that is responsible for receiving messages from other daemons
- Added site multicast for Intrusion-Tolerant Messaging (Priority and Reliable Messaging):
- Messages sent with a specific destination port, when received at the destination daemon, are delivered to all clients of that daemon that have joined the multicast group
- Added session option to specify behavior when the buffer between a daemon and its client fills:
- Default option maintains connection but drops new packets that do not fit
- Other option closes the client connection when the buffer overflows
- Initial separation of K-paths from Intrusion-Tolerant Messaging - *EXPERIMENTAL*
- Added high resolution timestamps to Alarm logging
- Some minor bug fixes, including:
- Specific erroneous client connections are now cleaned up and do not cause daemon to exit
- Multi-path routing now correctly delivers messages to local clients in a multicast group even if the destination daemon cannot be reached
Version 5.1 - May 17, 2016
- Added support for Inter-Process Communication (UNIX Domain Sockets)
- Daemon binds to default filepath, non-default path can be specified on cmd line or in config file
- Changes to libspines to support AF_UNIX connections (on UNIX architectures)
- Testprogs and Controlprogs updated - IPC is new default, can connect via non-default path or AF_INET
- Testprogs: sp_bflooder, sp_uflooder, sp_tflooder, port2spines, spines2port, sp_ping, sp_xcast
- Controlprogs: setlink
- New design (and improvements) for session blocking in Reliable Messaging
- Added DL_send_connected() to send scatter gather using send_msg function for connection-based AF_INET and AF_UNIX sockets
- Added new session semantics (chosen by application) needed for IT reliable communication:
- RELIABLE_STREAM: Block session when flow buffers become full toward any destination
- RELIABLE_DGRAM_NO_BACKPRESSURE: Silently drop messages that do not fit, no blocking
     *Experimental* (May be subject to change)
- RELIABLE_DGRAM_WITH_BACKPRESSURE: Explicit feedback indicating which messages are accepted, no blocking
     *Experimental* (Not in open source yet)
- Renamed both of the intrusion-tolerant messaging semantics to more accurately match behavior:
- Changed BEST_EFFORT_FLOOD_ROUTING to IT_PRIORITY_ROUTING
- Changed RELIABLE_FLOOD_ROUTING TO IT_RELIABLE_ROUTING
- Several minor bug fixes, including:
- Reliable Messaging - fairness in the presence of a certain attack
- Large bitmask support - for bitmasks larger than 64 bits
- Reliable Session - now correctly uses the routing notion
Version 5.0 - January 26, 2015
- Added support for a configuration file (spines.conf) to specify:
- Network topology (nodes and edges)
- Parameters for new link and dissemination protocols
- Cryptographic options
- Intrusion Tolerance Mode - a new ability to harden Spines against potential insider attacks
- Support for multiple dissemination methods - each message now has both a dissemination and a link protocol
- New link protocol - Intrusion Tolerant Link
- New dissemination protocol - Priority Messaging with Source Fairness
- New dissemination protocol - Reliable Messaging with Flow Fairness
- Source-based routing, including node disjoint paths, using a bitmask per message
- Integrated with OpenSSL functions to provide cryptographic authentication in the new protocols above
- New test program (sp_bflooder), which functionally resembles sp_uflooder, but supports the new protocols above
Version 4.0 - September, 2013
- Upgraded to use StdUtil 1.1.3
- Reorganized to use LibSpreadUtil 4.2.0
- Common forwarding and routing logic factored out and concentrated in route.c
- Minor improvements in state flood and hello protocols
- Minor improvements in build system; general code clean up
- Node, Interface, Leg and Edge ID's virtualized -- dissociated from network addresses (prepare for IPv6)
- Experimental SSL code removed
Version 3.2 - May 15, 2009
- New package directory layout and organization.
- Mac OS X updated support.
- Win32 updated support.
- Updated Wireless code, so it is not compiled in unless -WIRELESS_SUPPORT option is enabled on the configure command line.
- SPINES_IP_TTL and SPINES_IP_MULTICAST_TTL time-to-live options on spines_setsockopt now available and supported.
- Added errno error reporting on client library errors.
- Upgraded to use StdUtil 1.1
- Several minor bug fixes.
Version 3.1 - Mar 21, 2008
- Added spines_setsockopt() options to control Time-to-Live (ttl) value on overlay unicast/multicast/anycast packets.
- Added ability to retrieve overlay membership information through a spines_ioctl() call.
- Added ability to retrieve overlay network topology/distance/cost information through a spines_ioctl() call.
- Added ability to retrieve average network cost from any node to the members of a group through a spines_ioctl() call.
- Added wireless monitor capability to retrieve signal strength (RSSI) and other parameters in wireless environments.
- Several minor bug fixes.
Version 3.0 - May 31, 2007
- Added neighbor autodiscovery through IP-Multicast or broadcast.
- Added unicast kernel routing between daemons.
- Added kernel routing based on Multicast/Anycast group membership.
- Added Anycast group membership service.
- Added wireless option for setting different timers and behavior on wireless environments.
- Added multicast Loop option for setsockopt.
- Added endianess support for cross-platform deployments.
- Added SSL-based security on point-to-point communication between the daemons (experimental).
Version 2.2 - Dec 20, 2005
- Added support for creating custom topologies with dynamic link characteristics.
- Several minor bug fixes.
Version 2.1 - Jan 18, 2005
- Changed the API to closely resemble the regular socket interface.
- Added option for UDP communication between the applications and overlay nodes.
- New Real-Time link and routing protocol for use in time-sensitive applications such as VoIP.
- Added support for large data packets, up to 64KB for datagram communication and virtually
unlimited for stream-based communication.
- Added end-to-end application flow control for stream-based communication.
Version 2.0 - Dec 14, 2003
- Added Multicast capabilities.
- Routing based on distance, latency or loss rate of the links.
- Stream based fairness option for the reliable link congestion control.
- Added setloss monitor command for setting loss and burstiness on overlay links.
Version 1.0 - Jan 17, 2003