next up previous
Next: Scaling the Network Up: Performance Results Previous: Performance of Remote

Scaling the Input

To understand the impact of the working set size on the paging policy, we measure the execution time of one of our applications ( FFT), as a function of its input size. The completion time of FFT both under REMOTE_MEMORY and under DISK is plotted in figure 3. We see that as soon as the working set size exceeds 20 Mbytes, the paging starts, and the completion time of the application rises sharply. Most users would not be willing to tolerate such a high overhead in order to run an application that does not fit in main memory. Fortunately, REMOTE_MEMORY reduces this overhead by more than a factor of two.

  
Figure: Performance of FFT as a function of input size when either DISK, or REMOTE_MEMORY are used as backing store.



Evangelos Markatos
Fri Mar 24 14:41:51 EET 1995