Join Product Skew [cont'd.]

• The case of *100 Join *1000 tested with different number of processors per node is presented in table 5.

No. of Virt Procs		  1 		   5 		   10 		   20 		  30 		  60 

Exec. (Seconds) 		  147.2		   95.3		     64.0		    54.0		  51.8		  49.7

• Observations :
  (i) The processor scheduling algorithm was uniformly worse so its data has been omitted .
  (ii) There is a clear trend that more the virtual processors the better performance. The reason is that the '1's are being distributed over more and more z actual processors .