DISTCACHE

Section: distcache (8)
Updated: 2004.03.23
Index Return to Main Contents

NAME

distcache - Distributed session caching

DESCRIPTION

The distcache architecture provides a protocol and set of accompanying tools to allow applications, and indeed machines, to share session state between them by way of a network service.

The primary use of distcache right now is SSL/TLS session caching. This allows SSL/TLS servers (eg. a secure Apache web server providing HTTPS support) to use a centralised session cache, i.e any server may resume SSL/TLS sessions negotiated by any other server on the network. The advantages to this approach include increased freedom of mechanisms for load-balancing.

Existing SSL/TLS load-balancing solutions

Many load-balancers attempt to route incoming connections to servers based on remembering the last mapping from the same source network address. Others, called ``SSL sticky'' load-balancers, attempt to parse SSL/TLS session ids from handshake messages and so map future session-resume attempts. Both methods have serious weaknesses - the former is generally confused by any form of network address translation (eg. when clients are behind masquerading gateways), and the latter is confused by any SSL/TLS renegotiations. Moreover both are stateful and a potential bottleneck, because there is no obvious way to scale the architecture to multiple load-balancers.

Arbitrary SSL/TLS load-balancing with distcache

There is no need nor motivation to route incoming connections to ``the same server'' to improve the chances for SSL/TLS session resumption, and indeed doing so defeats the point of load-balancing (which is to balance according load or availability). The use of distcache is to ensure that all servers share the same ``cache'' and so can respond to SSL/TLS session resume requests irrespective of where the previous SSL/TLS connection from the same client was mapped to.

ain't just for SSL/TLS ...

Future versions of distcache will expand on the protocol and should provide for a variety of ``shared-state'' uses besides SSL/TLS session caching. The possibilities include application state caching, network-based shared virtual memory, etc.

TOOLS

dc_server: Runs a cache server listening on a configurable network address. See dc_server(1).
dc_client: Runs a local client proxy. From the point of view of applications this behaves like dc_server, but manages multiplexing application requests to/from a cache server over a single persistent connection. See dc_client(1).
dc_test: Sends a (configurable) barrage of session caching requests to a given network address using the distcache protocol. Useful for testing correctness of an installation as well as benchmarking. Can be used directly against an instance of dc_server or against a dc_client proxy. See dc_test(1).
dc_snoop: A transparent proxy tool supporting the distcache protocol that can be used to monitor cache operation requests and responses between any two end-points (eg. between an application and dc_client, or between dc_client and dc_server). See dc_snoop(1).

APIS

The comments below provide a short summary of the APIs available in distcache. To view more details, consult the section 2 man pages these summaries refer to. If you are using a packaged version of distcache, you may need to ensure that a corresponding ``devel'' package is installed as the libraries, headers, and API documentation is often packaged independantly of the user tools.

libnal

This is the underlying Network Abstraction Library (hence ``NAL'') used by the distcache libraries and tools. libnal uses non-blocking sockets, with an addressing abstraction that allows tools to transparently work over unix domain sockets or TCP/IPv4 sockets by a change of address text. For this reason, all the distcache tools can have their ``-listen'' and ``-connect'' switches set to work over either kind of transport.

libnal defines various object types;

NAL_ADDRESS (see NAL_ADDRESS_new(2)): The addressing abstraction converts to and from text representations, indicates whether given addresses are valid for listening on, connecting to, or both.
NAL_CONNECTION (see NAL_CONNECTION_new(2)): This encapsulates a network connection that can be used for sending and receiving arbitrary binary data.
NAL_LISTENER (see NAL_LISTENER_new(2)): This encapsulates a listening socket that can be used to accept incoming connection requests on a configured address, creating a NAL_CONNECTION wrapper for each accepted connection.
NAL_SELECTOR (see NAL_SELECTOR_new(2)): This provides an object that can be prepared with various NAL_LISTENER and NAL_CONNECTION objects, and then can block waiting for network activity up to some configurable limit of time. This is the basis of non-blocking I/O and is an encapsulation of the traditional select(2) function.
NAL_BUFFER (see NAL_BUFFER_new(2)): This abstraction implements a FIFO data array and is used primarily for representing the read and send parts of a NAL_CONNECTION object.

There are also some helper functions to assist in serialising data, particularly with respect to putting integral data into network byte order (allowing interoperability between platforms with differing byte-order). These functions are documented in NAL_decode_uint32(2).

libdistcache

There are two APIs implemented by the libdistcache library;

distcache/dc_plug.h: This header provides the DC_PLUG abstraction. This encapsulates a connection and implements the distcache protocol and various functions for manipulating the reading and writing of distcache messages (requests or responses). This abstraction can support client and server implementations of the distcache protocol and supports asynchronous behaviour by interacting with libnal's NAL_SELECTOR type. For more information, see DC_PLUG_new(2).
distcache/dc_client.h: This header declares a higher-level (and much simpler) API than dc_plug.h, and is useful in applications that want API functions that ``do cache operations''. The API is blocking, and provides simplistic ``add'', ``remove'', and ``get'' functions that only return once the full request/response cycle is complete or an error has occured. This is the API used to add distcache support to applications like Apache, stunnel, etc. For more information, see DC_CTX_new(2).

libdistcacheserver

This header declares an API for implementing a session cache supporting the distcache protocol. It is primarily intended for environments that wish to implement an alternative method for session storage. As with elements of libdistcache, this API is likely to be undergoing some important restructuring and enhancements. Please consider subscribing to the distcache mail list and/or monitoring CVS, this gives you an opportunity to influence ongoing development and be less surprised at changes the turn up in future versions. For more information, see DC_SERVER_new(2).

LICENSE

The distcache toolkit, including the libnal network abstraction library that comes bundled with it, is distributed under the LGPL license (``Library GNU Public License'') and you should have received a copy of this license with this software and its documents.

BUGS

Quite possibly. In particular, portability has not been tested under many platforms as the current developers have limited OS resources. Feedback, access to alternative platforms, bug-reports, and questions are all welcome - please go to the distcache website and subscribe to the distcache-users mail list.

AUTHOR

This toolkit was designed and implemented by Geoff Thorpe for Cryptographic Appliances Incorporated. Since the project was released into open source, it has a home page and a project environment where development, mailing lists, and releases are organised. For problems with the software or this man page please check for new releases at the project web-site below, mail the users mailing list described there, or contact the author at geoff@geoffthorpe.net.

Home Page: http://www.distcache.org