1) Look up Mark Harwood and Lucene.  He used Lucene to distribute
searches some time ago (1> year ago) and provided some nice sequential
UML diagrams with notes.  I believe I saw something similar from Kevin
Burton, also a long time ago, and of course Nutch system is designed
with a cluster of search servers in mind.

2) I think that depends on your needs.  For everything I used Lucene
for, having a duplicate copy of an index was always sufficient, as I
could always scp it to a live machine in a few seconds or minutes.

3) I would not call it pointless, although with time the FS caches
'warm up' and the search performance may approach that of
RAMDirectory-backed indices.  Converting a FSDirectory based index to a
RAMDirectory one is so easy, that there is no point in trying to
optimize early.  I would keep my code simple and just try using
FSDirectory first, until I start having performance issues.

Otis


--- jt oob <[EMAIL PROTECTED]> wrote:
> Hi,
> 
> I posted this message a few weeks back, its the first time i haven't
> had a reply from the list! If anyone could comment on the 3 questions
> I
> raised that would be great.
> 
> Original:
> 
> # From: jt oob
> # Subject: (Distributed) Search system designs
> # Date: Fri, 14 May 2004 07:20:08 -0700
> 
>
http://www.mail-archive.com/[EMAIL PROTECTED]/msg07388.html
> 
> Thanks,
> 
> jt
> 
> 
> 
>       
>       
>               
> ____________________________________________________________
> Yahoo! Messenger - Communicate instantly..."Ping" 
> your friends today! Download Messenger Now 
> http://uk.messenger.yahoo.com/download/index.html
> 
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: [EMAIL PROTECTED]
> For additional commands, e-mail: [EMAIL PROTECTED]
> 


---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]

Reply via email to