Progress on the MSc project

The various process states, displayed in a sta...
Image via Wikipedia

Had a letter from Birkbeck today telling me I had passed all my exams – so the issue now is finishing the MSc project and so getting the degree (I suppose, in theory at least as I have yet to be formally awarded it, I could now claim to have reached post graduate diploma level).

Right now, on the project, I am testing small kernel patches to see if a localised page replacement policy makes any noticeable difference for systems that are, or are in danger of, thrashing (ie spending nearly all their time waiting for paging to and from disk as opposed to doing any real computing).

The first patch I tried, forcing the largest and a large (not necessarily second-largest) process to write any dirty pages to disk had no noticeable effect – having thought about it a bit more, I now realise is unlikely that file backed pages are going to be dirty in this way for many processes, so actually all this code is likely to have done little but degrade performance.

What I really need is an aggressive rifle through the biggest processes page stack, essentially accelerating the hands of the CLOCK (of the CLOCK page replacement algorithm)  for this process (or, diminishing the handspread, to use a term often seen in this context) compared to those in general. So that’s next.

About these ads