glen herrmannsfeldt <g...@ugcs.caltech.edu> writes: > It seems to me that adaptive algorithms are more likely to sync > to each other when nested. But how about one that examines every > Nth page, (hopefully N is prime), such that they won't be the > exact same pages. Or even using a more random path, such as from > a CRC polynomial. So the path through the pages will be different, > and so different approximately LRU pages will be selected. > > Never having tried this, those are the ones I think up.
re: http://www.garlic.com/~lynn/2012b.html#98 5 Byte Device Addresses? http://www.garlic.com/~lynn/2012b.html#100 5 Byte Device Addresses? http://www.garlic.com/~lynn/2012c.html#16 5 Byte Device Addresses? http://www.garlic.com/~lynn/2012c.html#17 5 Byte Device Addresses? http://www.garlic.com/~lynn/2012c.html#27 5 Byte Device Addresses? http://www.garlic.com/~lynn/2012c.html#28 5 Byte Device Addresses? http://www.garlic.com/~lynn/2012c.html#29 5 Byte Device Addresses? that is strickly deterministic ... these are all approximate LRU selection for replacement. The theory behind choosing least recently used pages is that they have shown to be the least probable being used in the future. if the VM system is choosing virtual machine pages for replacement based on least recently used ... and the guest MVS system is looking for pages have been also least recently used ... they both will tend to concentrate on selecting from the same subset of pages ... the guest MVS selecting their least recently used virtual machines and the VM system selecting the MVS guest virtual machine pages that the corresponding MVS virtual pages occupy. That significantly increases the probability that the page the guest MVS selects for replacement and the corresponding virtual machine page to use ... that guest virtual page has also been selected by the VM system for replacement and removal from real memory. Running a least recently used replacement algorithm under a least recently used replacement algorithm violates the assumption that the least recently used page is the least likely to be used in the future. They don't have to be strickly in sync ... but it will drastically increase the probability that there is double paging ... aka the virtual machine page that the MVS system wants to start using is a page that the VM system has removed from memory. As I previously mentioned, something similar happens with a large DBMS cache being managed by least recently used ... running in a virtual memory operating system. It is one of the reasons that virtual memory operating systems tend to have ways of biasing against selecting large DBMS cache pages (because they useage patterns tend to violate the assumption that the least recently used page will be the least probable page to be used in the future). misc. past posts mentioning virtual memory replacement http://www.garlic.com/~lynn/subtopic.html#clock old email mentioning various aspects page replacement ... including work related to "big page", full-track page transfers implementation also resulted in tweaks that resulted in underminning least recently used (corresponding to something similar done for the original SVS implementation that continued well into MVS releases) http://www.garlic.com/~lynn/lhwemail.html#globallru -- virtualization experience starting Jan1968, online at home since Mar1970 ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: INFO IBM-MAIN