Henri, Thanks again for another prompt and extremely helpful response.
Henri Asseily wrote: > > I humbly repeat my earlier suggestion to use DBIx::HA. I'm going to try it again today, using the same benchmark system. I'll report back the results. I started looking at DBD::Multi again because it load balancing built-in, even though it should be easy to do it with DBIx::HA, as you've described. I also realized that at least initially, I won't really be doing load balancing-- I'll just be sending some SELECTs to a slave to reduce the load on the master, which must keep accepting all the write activity. I was also put off by the discovery that it doesn't support "do()". Later, as I thought about it more, the obvious became clear: I won't ever be using "do()" with SELECT statements, so that didn't matter. > ResourcePool >> http://search.cpan.org/dist/ResourcePool/ > > ResourcePool is a generic resource pooling framework. Considering that > Apache::DBI already exists, if you use it in conjunction with > Apache/mod_perl and iThreads you'll get the exact same functionality. > You don't gain much by using ResourcePool for DBI resources. > Also ResourcePool only works in iThreads, so no cross-process pooling. > >> SQL-Relay >> http://sqlrelay.sourceforge.net/ > > SQL-Relay is yet another similar approach to high availability. It > does load balancing, failover and connection pooling and caching. It > works by having the user connect to SQL-Relay which then calls the > database, grabs results and returns them to the user. It attempts to > be as close to a direct database call as possible, so it needs drivers > for every database engine, and it needs to map as many API calls as > possible, including the low-level calls. In effect, it's like DBI but > in C, and has support for query caching. In order to do connection > pooling, it uses many listener processes (those that talk to the > client) and many connections processes (those that talk to the > database), and has all these creatures talking to each other through > shared memory. If you're comfortable with such an approach, go ahead. > It does seem that the Sybase driver is rather immature, but the Oracle > driver and probably the MySQL one are seeing heavy use and > development. SQL-Relay is very powerful, and it may work well enough > for you. These are most useful descriptions of these tools that I've found. SQL-Relay does sound like a possibility for me, but I'll stick with a simpler solution such as DBIx::HA if it works. Mark
