[Alpine-info] sort_thread_flatten() performance

[Dimitrios Apostolou]

On Thu, 10 Jan 2008, Mark Crispin wrote:

> On Fri, 11 Jan 2008, Dimitrios Apostolou wrote:

>> I just tested alpine on OpenBSD with the for loop commented out. It seems 

>> *much* faster: it only needs ~5sec after it the server replies, great 

>> improvement. So it seems that recursion isn't that bad, as I presumed.

>> 

>> Memory usage is still ~60MB, I guess this is relevant to the data 

>> structures alpine uses. Anyway, I think 60MB is not bad for such a large 

>> mailbox.

>

> I think that I know where the 60MB is coming from.  The server does not 

> support threading, so the c-client library has to do it locally.  This, in 

> turn, requires fetching all the data needed to thread.  That data, once 

> fetched, is cached.

>

> On a server that supports threading, that data isn't fetched until you do 

> something such as scroll to that message in the index.


I am pretty sure that dovecot supports threading, and that it is being 
used as it should. I have checked the debug file and I get a proper 
response to "THREAD REFERENCES US-ASCII ALL" command. And I'm 100% sure 
that alpine doesn't cache all headers locally, but requests the ones that 
are viewable on the screen as I scroll.

I fact, you previously mentioned a memory usage of 30MB for ~50K messages. 
I get ~40MB for ~200K messages. And after threading memory usage exceeds 
60MB. I believe you'll see the same numbers in a larger mailbox.


Thanks, 
Dimitris



>

> -- Mark --

>

> http://staff.washington.edu/mrc

> Science does not emerge from voting, party politics, or public debate.

> Si vis pacem, para bellum.

>