performance issue (imap spool on san)

[Andrew Findlay]

On Fri, Jul 28, 2006 at 10:37:43AM +0100, Alain Williams wrote:

> Might it not be better to have Cyrus 'learn' what header lines are needed,
> rather than just bloating the list with more headers. The set of headers
> would needed to be dynamically changable.  The points are:
> 
> 1) different IMAP clients want different sets of headers. The same IMAP
>    client at different releases might change the set requested.
> 2) most individual sites run only 2 or 3 different IMAP clients, why get
>    Cyrus to collect the headers that the IMAP clients at that site don't want.
> 3) most system admins don't have the skills/inclination/... to optimise
>    the set of headers cached.

The problem with the learning approach is that the index is no use at
all unless it contains the same headers from every message, and it is
built at message delivery time rather than when the message is read.

I suppose some sort of 'lazy indexing' could add headers to the index
on demand, but the code changes might be significant.

Headers are not usually very large, so I would be more inclined to
the idea that the index should store every header (perhaps with a
blocklist to avoid things like Received:)

Andrew
-- 
-----------------------------------------------------------------------
|                 From Andrew Findlay, Skills 1st Ltd                 |
| Consultant in large-scale systems, networks, and directory services |
|     http://www.skills-1st.co.uk/                +44 1628 782565     |
-----------------------------------------------------------------------