Re: New functionality proposed

[Howard Chu <hyc@symas.com>]

It took some effort to make sense of this email, but the idea seems 
interesting. I would start by looking at the ProxyCache overlay since 
intelligent caching may reduce your reliance on the other aspects of 
your idea. Also, as LDAP is a session-oriented protocol, your use of the 
word "broadcast" is suspect. A decent design in a bandwidth-limited 
environment would choose a single target server from the pool of 
available servers and make a single request to that server; broadcasting 
a request to all available servers is only possible when all of the 
servers are on the same LAN as the querier and even then it's of 
questionable merit.

Next I would look at extending back-ldap to allow runtime-specifiable 
target servers instead of a single statically configured target. How you 
manage your rules to select which server to use can be solved as a 
completely independent issue.

Bengt Hammarlin wrote:<>

Hi guys, I am proposing to develop new functionality for "intelligent" 
routing of requests to a distributed LDAP database.
This would be done to better utilize network and server resources, 
especially where infrastructure might be insufficient.
The assumptions made for this are
  The database is relatively large
  The database is used on several systems geographically distant
   Data can be logically localized
   A vast majority of directory lookups are local
  Network and/or server resources can be exhausted if every request is 
broadcast to all servers

Based on that I am suggesting a design along these lines.

A new backend is done based on current ldap or meta backends
This backend would have a base functionality that first sends the 
request to the defined local LDAP server and if it fails the request is 
broadcast to the remaining servers in the setup. There would be 
configurable options
  A rule based request analyzer that would based on rules route the 
request to a specific server (assuming sufficient data present in the 
request)
  Caching that would store key data locally to determine witch server 
to route the request to
  Key data replication to determine where to route requests

I  would appreciate thoughts and suggestions on this plus any 
documentation on how to develop a backend and/or on suggested backends 
to use as a starting point. Bengt Hammarlin   = ;Office +46 60 16 10 86 
Software Designer = ; Mobile= =20 +46 73 325 8506 Mobeon AB

--
 -- Howard Chu
 Chief Architect, Symas Corp.       Director, Highland Sun
 http://www.symas.com               http://highlandsun.com/hyc
 Symas: Premier OpenSource Development and Support