Date: Tue, 10 Dec 1996 03:38:07 GMT Server: NCSA/1.4.2 Content-type: text/html Last-modified: Sun, 04 Jun 1995 23:39:43 GMT Content-length: 4634 Implementation of Dynamic Superpage Management

Implementation of Dynamic Superpage Management

We're looking for someone to help us implement dynamic superpage management as a quals or masters project. Send me (romer@cs.washington.edu) mail if you're interested.

Background

In recent work, Wayne Ohlrich, Anna Karlin, Brian Bershad, and I have explored policies that monitor application memory reference patterns in order to identify and resolve TLB performance problems. Poor TLB performance results when the TLB is too small to cover the current application's working set. Several modern architectures support superpages: pages whose size is a multiple of the system's base page size. On these systems TLB performance can be improved by using larger pages, but at the cost of wasted memory due to internal fragmentation. We explored several policies that adapt the page size dynamically to different regions of an application's address space, constructing superpages by copying the component pages to a contiguous region of memory. We developed a policy that monitors TLB misses, and balances the potential benefit of having a superpage (a reduction in future TLB misses) against the cost of constructing the superpage (an in-memory copy). By constructing superpages only when and where TLB miss patterns warrant, this policy attains the TLB performance of large pages without their internal fragmentation.

See the paper Memory Systems Research at UW