[Date Prev][Date Next] [Chronological] [Thread] [Top]

Re: Connection Pooling



--On Tuesday, April 15, 2003 8:24 AM -0400 Tod Thomas <tthomas@chubb.com> wrote:

From an application's perspective is it really necessary?  Because LDAP
connections are so transient, unless there are a fairly large amount of
requests being served I would suspect that LDAP's read optimizations
would make this unnecessary.  Comments?


Tod,

I disagree. There can be a world of difference in the query rate that varies on a large number of things. In Stanford's setup, where we use K5 authentication via Cyrus-SASL and GSSAPI, we get a query rate of about 66 queries/second using ldapsearch with 18 hosts querying. Using Perl to make the connection, we can get 44 queries/second with a single host doing serialized binds. Using connection pooling, we can get from 220 queries/second to 460 queries/second from a single host, depending on how the search is structured. For us, with the large number of queries being made against our servers, utilizing connection pooling is a step we intend to pursue. I'm also not quite sure what you mean by connections being transient. That they mainly come from unknown sources? If that is what you mean, I would say that depends on your setup. Our largest client of data at this time is our email service, which runs an ldap query to look up the delivery address of all incoming email. With over 54 million queries a month, that service is a constant load on our servers, and is not something I would qualify as transient.

Regards,
Quanah


--
Quanah Gibson-Mount
Senior Systems Administrator
ITSS/TSS/Computing Systems
Stanford University
GnuPG Public Key: http://www.stanford.edu/~quanah/pgp.html