seen file corruption

[Richard Houston]

Hmmm I see your point. Maybe not such a good idea.
Thanks for setting me straight.

Rich

On Fri, 2003-06-13 at 10:19, Rob Siemborski wrote:
> On Fri, 13 Jun 2003, Richard Houston wrote:
> 
> > It would be very cool to have the actual messages stored in database as
> > well. The dbmail project is doing this in Mysql or Postgresql. Imagine
> > the maturity of Cyrus with all the benefits of storage and backup in a
> > database. Would be cool.
> 
> Except that the benefits of Cyrus are lost, for the most part, if you
> change the mail store.  The parser APIs that lie on top of the mail store
> are almost trivial to implement compared to the data storage (and
> retrieval) mechanisms themselves.
> 
> Cyrus gets its high performance from clever indexing and caching of
> information in each mailbox.  If you gut all this code and replace it with
> a database, you've removed the primary benefit!
> 
> Additionally, you've probably lost some performance because the speed of
> Cyrus's store (optimized for IMAP) is going to be hard to beat with a
> generic database.  Not impossible, but its really not something we're
> interested in doing for the foreseeable future.
> 
> As it is, I have trouble understanding why anyone would want to make a
> cyrusdb out of a relational database.  Cyrusdb provides simple key/value
> pairs, where a relational database is providing a lot more, and has more
> overhead.  If you want to be able to access this data (internal to cyrus)
> from another application, you're asking for serious trouble.
> 
> -Rob
> 
> -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
> Rob Siemborski * Andrew Systems Group * Cyert Hall 207 * 412-268-7456
> Research Systems Programmer * /usr/contributed Gatekeeper
> 
> 
> 

On Fri, 2003-06-13 at 10:19, Rob Siemborski wrote:
> On Fri, 13 Jun 2003, Richard Houston wrote:
> 
> > It would be very cool to have the actual messages stored in database as
> > well. The dbmail project is doing this in Mysql or Postgresql. Imagine
> > the maturity of Cyrus with all the benefits of storage and backup in a
> > database. Would be cool.
> 
> Except that the benefits of Cyrus are lost, for the most part, if you
> change the mail store.  The parser APIs that lie on top of the mail store
> are almost trivial to implement compared to the data storage (and
> retrieval) mechanisms themselves.
> 
> Cyrus gets its high performance from clever indexing and caching of
> information in each mailbox.  If you gut all this code and replace it with
> a database, you've removed the primary benefit!
> 
> Additionally, you've probably lost some performance because the speed of
> Cyrus's store (optimized for IMAP) is going to be hard to beat with a
> generic database.  Not impossible, but its really not something we're
> interested in doing for the foreseeable future.
> 
> As it is, I have trouble understanding why anyone would want to make a
> cyrusdb out of a relational database.  Cyrusdb provides simple key/value
> pairs, where a relational database is providing a lot more, and has more
> overhead.  If you want to be able to access this data (internal to cyrus)
> from another application, you're asking for serious trouble.
> 
> -Rob
> 
> -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
> Rob Siemborski * Andrew Systems Group * Cyert Hall 207 * 412-268-7456
> Research Systems Programmer * /usr/contributed Gatekeeper
> 
> 
>