[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Client update protocol



Randy Turner wrote:


Hi,


Thanks for the reply. Regarding your earlier comments...

I might have 30,000 LDAP clients that want to know if a particular LCUP context has changed. They might "subscribe" to this particular event so they get notified. Notification methods such as that suggested by persistent search (and variants) or polling (connect/bind/search) would be sub-optimal with this many clients potentially involved. The LCUP sync mechanism will work great...I just need a client to be "tapped on the shoulder" when it should connect/bind and issue the LCUP search method. The same entity that is notified, will be the same entity that issues the special LCUP context search.

Then, will you have 30,000 clients making LCUP search requests at the same time, if they are all interested in the same event?


At an IETF meeting a couple of years ago I had a discussion about such a technique (before I actually had a need for it) during early discussions about CLDAP and how this type of scenario could be included in the problems addressed by CLDAP.

This seems like a perfect application for multicast. Making 30000 simultaneous or sequential TCP/IP connections is sub optimal.


Your suggestion about a "middleware" entity that maintains a persistent LCUP search sounds like it would be a good start, looking "northbound" to the source DIT. Looking back towards the clients, there would still be a need for clients to register their interest, as well as the actual lightweight delivery mechanism.

If there had been an extensible "trap/inform" group in the MADMAN MIB, then that might be a potential solutions as well. Not perfect, but a step in the right direction. Ultimately, the solution will probably involve work going on outside the LDAP WG, again I'm just trying to find anyone with a problem domain similar to this.

I don't remember the exact names of the projects regarding event notification systems...but if you do a google (or other) search for "scalable event notification", this will show you a set of starting points that I used for my own research. Most of the recent work has been in academia, but I believe there are implementations of these systems available, as well as some organizations looking to go commercial with similar technology.

Thanks again for the reply,
Randy


On Monday, February 10, 2003, at 01:26 PM, Rich Megginson wrote:


Randy Turner wrote:

Hello all,

I recently read over the LCUP document and this text looks very good for DIT synchronization. What I am looking for is some type of "event" mechanism to trigger a client to connect/bind and initiate LCUP. This would require the directory server to have knowledge of "who" is interested in being notified of DIT updates. I may have 500,000 clients for a directory, and I would prefer not to have to have the clients poll the server. If a particular LCUP context is modified, I would like some type of async notification sent to interested parties. Most of the work of actually synchronizing the DIT is covered by the LCUP document, I just need the clients to be notified "when" to perform LCUP. Ideally, the trigger mechanism would be some type of lightweight, connectionless notification, similar to an SNMP trap but more specific.


How many "interested parties" will there be? Will the "interested parties" actually initiate an LCUP request?

A standard LDAP server is not designed to initiate connections to other servers or clients (except in the case of replication, but that is not a standard yet, nor would it likely serve your purpose). Some LDAP servers do support callbacks or triggers that usually involve having to write your own plugin or some other mechanism.

This sounds like the sort of thing that message bus/queue architectures are trying to solve. You could have an LDAP client that issues an LCUP persistent search against an LDAP server, and the client would put a message on the bus/queue every time it received an LCUP response. Other clients register with the bus/queue and receive

BTW, for those interested, we are still working on the LCUP 04 draft and hope to have something very soon.

There is activity going on with global, scalable event notification architectures that is not specific to any IETF work, and we are looking into this. However, I wanted to make sure I understand the taxonomy of any and all work related to solving this type of problem.


Can you provide more information about this work?

If there is work going on in this area in this, or another working group, I would appreciate any and all pointers, or other comments.

Thanks in advance,
Randy


<smime.p7s>



Attachment: smime.p7s
Description: S/MIME Cryptographic Signature