- From: Sergey Melnik <melnik@DB.Stanford.EDU>
- Date: Tue, 22 Aug 2000 05:37:56 -0700 (PDT)
- To: Guha <guha@guha.com>, Jeen Broekstra <jbroeks@cs.vu.nl>
- cc: www-rdf-interest@w3.org
Guha, Jeen, the only change I made on Solaris was to include the socket library for linking, i.e. in makefile, replace LIBS = -ldb -lpthread by LIBS = -ldb -lpthread -lsocket Guha, query optimization for a homogeneous triple store might need some novel techniques. A traditional DBMS can hardly maintain useful statistics across all value combinations. Therefore, bad query plans are often generated. I have some ideas, let's talk about it, I'll be back next week. Sergey On Fri, 18 Aug 2000, Guha wrote: > Hi Sergey, > > Did you have to make any chage to the code to get it running > on Solaris? > > I have plans of supporting reification. Just haven't done it yet. > > The indexes are currently built on b-trees provided by Sleepycat. > I am working on a "nested hash" structure which is more appropriate > for graphs. The "join algorithm" is fairly straightforward and > only uses conjunct ordering for optimization. I plan to implement > a lot of the optimizations used in cyc, lore, etc. > > Thanks! Do you have any suggestions, advice, etc? We should talk > about this sometime. > > Guha
Received on Tuesday, 22 August 2000 08:43:01 UTC