On Nov 2, 2006, at 7:34 AM, Leland, Robert wrote: >/ In our three machine cluster the load distribution is: />/ server 1 is handling about 50% (Dell 2850) />/ Server 2 is handling about 25% (Dell 2650) />/ Server 3 is handling about 25% (Dell 2650) />/ />/ DNS is pointing to Server 1 & Server 2, in a round robin fashion. />/ />/ Is the uneven distribution because resin is detecting that server 1 is />/ more powerful ? / Possibly.
Currently, Resin distributes a new connection to the server with the least number of active connections. (With 3.1, we might add a CPU usage component to that cost calculation.) If one server is faster, it will complete connections faster and therefore be ready for a new connection before the other servers. So, if that server is twice as fast at completing connections, it should get twice the number of connections. -- Scott I'm experiencing this as well with the latest resin pro snapshot and mod_caucho. I have 4 identical machines and the first server (index="1") in line gets double amount of requests compared to the rest, bogging the machine down. Interestingly if I bring the first server down the remaining 3 nodes will balance just perfectly. I've used the same machines with the same ordering with Resin2 and it's mod_caucho without this problem for quite some time. Apache 2.0.46 with both cases. --Jaakko // _______________________________________________ resin-interest mailing list [email protected] http://maillist.caucho.com/mailman/listinfo/resin-interest
