No announcement yet.

Odd behaviour for a two-machines cluster

  • Filter
  • Time
  • Show
Clear All
new posts

  • Odd behaviour for a two-machines cluster

    Hi everyone

    I'm playing with the new VoltDB 4 on a cluster of two computers and I've been experiencing some trouble about the performance. I wrote a simple test program which just calls a remote procedure which in turn executes a simple INSERT, nothing else. The computers I've been working on are connected by a gigabit switch. The db instances are configured as K=0, 4 partitions per host.

    When I run the test on the first server, I get a speed of about 280k ops/sec; on the second one (which is a slower test machine) something around 140k ops/sec. When I cluster the two machines, the performance drops to 220k ops/sec, while I'm expecting something around a linear sum of the two speeds, no less than 300k ops/sec. What could be the bottleneck here? I checked memory, cpu and network bandwidth and they look ok on both machines.

    I also tried in the last week other experiments with, for example, a physical and a virtual machine, or two virtual machines, and in each case I saw that, when I cluster a faster machine with a slower one, the combined performance is not the sum but instead something like the average of the two.

    Can you imagine an explanation to this odd behaviour? Maybe I did something wrong with the configuration of the db instances?