From: Geoff H. <ghu...@ws...> - 2002-05-09 05:12:22
|
On Wednesday, May 8, 2002, at 09:47 PM, Neal Richter wrote: > We downloaded the latest copy of BDB from sleepycat. It contained > a win32-native db_config.h that made the Microsoft compiler tool-chain > happy. This is Berkeley DB 4.x? > We're considering just linking htdig.exe against the most recent > version of BDB to see if it goes away. It might not be a bad idea. It'll work for 3.1.x, but it's more complicated with 3.2. Although if you don't seem to have the problem with building mifluz on Windows and running the mifluz test suite, then 3.2 would be OK too. (Unfortunately, the BDB folks felt that the database compression used in mifluz wasn't useful enough for their customers and so they have not integrated the changes Loic coded in the Berkeley code back into their codebase.) > BDB is pretty complex code... it seems like the windows execution > path is very different from the Linux execution path.... so even a > comparative log of the writes doesn't work well. (different 'pagesize' > for DB files) IIRC, the Sleepycat folks did some benchmarking and the result is the modifications to the Windows environment to get as good a performance boost as possible. I don't think it's quite as fast as the Linux code, though. We looked at many of these issues when considering how to structure the word database for 3.2. Adding records is fast, making records longer is v. slow for all platforms and Berkeley DB types. -Geoff |