After a few days of work, sourceware’s new web search engine, mnogosearch is running.

Because of FSF political reasons, two separate installations / databases spider sourceware.org and gcc.gnu.org. The transition from htDig is not yet complete: it still runs sometimes, and there are still links to it from e.g., mail archive web pages. We’ll get to those soon.

The performance is so-so. mnogosearch uses a mysql database runing on a nearby dedicated computer for all its data. This was necessary because a local database incurred too great an additional load on the sourceware machine. This unfortunately means lots of tcp traffic going back and forth, especially during indexing. Search results are cached, so if you all look for “silly frank”, only the first person will get to pay the full waiting price.

Here are some interesting (?) statistics.

parameter	sourceware	gcc.gnu.org
mnogosearch version	3.2.25	3.2.25
mysql database size	11GB	18GB
number of documents	711000	974000
initial index run	24 hours	72 hours
refresh time	5 hours	8 hours

GIve it a try and enjoy. Let us know on the overseers mailing list what you think of it.