Re: [Gems-users] Change in cpu frequency - question

[Dan Gibson <degibson@xxxxxxxx>]

On Sun, Feb 7, 2010 at 11:07 AM, Pradeep Ramachandran <pramach2@xxxxxxxx> wrote:

[content removed for brevity]

But if I am simulating a server workload in a distributed client-server setting, wouldn't I want to make sure that I am having network communication (that is simulated by simics) in my measurement window?

The answer is: it depends. Specifically, it depends on what you're trying to study. If the important aspect of the client-server setting is the communication itself, then your infrastructure needs to model that communication in detail. On the other hand, if you're interested mostly in the client or server memory subsystem, then the actual communication between them is less important.

If your work requires a detailed model of the communication, then it could be that GEMS is not the right infrastructure. For one, GEMS models I/O rather badly. Secondly, it models the memory system in too much detail to allow a better I/O model -- it takes so incredibly long to simulate the memory system that adding realistic I/O events would require huge simulation times.

Regards,
Dan

-- 
http://www.cs.wisc.edu/~gibson [esc]:wq!