The experiment numbers for VXQuery on the weather data are getting more
consistent. Here is a look at the latest numbers. A few times are missing
and are highlighted in yellow. The recent upgraded to our cluster's network
switch has improved the q07 numbers. The new numbers listed are not bounded
by the network. All numbers are limited by the cpu.

The speed up graphs show how when the data size is kept the same, doubling
the machines half's the query time.

The batch scale up graph shows adding nodes with the same amount of data
does not increase the query time.

On the local experiments you will see reduced scaling after four
partitions, since these machines have four cores and the queries are cpu
bounded.

https://docs.google.com/spreadsheet/pub?key=0AsraYegikI-zdHhmQjZ3QlFrOVp6eV9mSGRfaEJfT0E&gid=15

Let me know what you think.

Preston

** Also let me know if this published google doc doesn't work for you.

Reply via email to