| 
|
Topic: memory management by working sets
topics >
computer science >
operating system >
data caching
Topic:
memory cache
Topic:
memory management by age
Topic:
memory management by paging
Topic:
memory management for programs and modules
Topic:
virtual memory
Summary
A working set contains the memory pages in active use. To avoid thrashing the working sets of running processes needs to be in memory. Typically, working sets change slowly with occassional abrut changes or phases. (cbb 12/07)
Subtopic: working sets and paging
QuoteRef: dennPJ_1978 ;;72 "The working set is usually defined as a collection of recently referenced pages of a program's virtual address space. QuoteRef: dennPJ_1978 ;;74 "Working set determination is best paging manager based on local performance of a program's page requests. QuoteRef: dennPJ1_1980 ;;extensive bibliography on working sets
Subtopic: multi-programming level
Quote: under paged, virtual memory, working-set strategies are better than FIFO disciplines [»oldeRR5_1985] Quote: if the client working set changes slowly, its description will remain in the caches, adding little or no extra disk activity [»redeDD2_1980] Quote: size working set--use an exact-fit cache of recently freed blocks; purge block after T allocations [»oldeRR5_1985]
Subtopic: thrashing
Quote: shed load to stay below the maximum paging rate [»hoarCA_1971]
Subtopic: reducing working set size
Quote: eliminate thrashing by monitoring the fault rate; keep damped counts of page faults, processor time, and I/O time [»hoarCA_1971] Quote: in a paging system, the only important thing is to avoid thrashing; otherwise spend all of the time waiting for the disk [»lampBW10_1983] Quote: least recently/frequently used (LRFU) block replacement policy; up to 30% better than LRU; uses linked list and priority heap based on threshold of recency and frequency value [»leeD9_1997]
Subtopic: CLOCK vs. working sets
Quote: reduce the hot working set by splitting structures into hot, frequently accessed portions and cold portions [»chilTM12_2000]
Subtopic: LRU and working sets
Quote: even though working sets are better, most modern multiprogramming system use the CLOCK memory policy [»dennPJ9_1980]
Subtopic: reuse distance
Quote: bounded locality interval: top elements of LRU with at least one additional reference; corresponds to phase [»hendDA7_1986] Quote: least recently/frequently used (LRFU) block replacement policy; up to 30% better than LRU; uses linked list and priority heap based on threshold of recency and frequency value [»leeD9_1997] Quote: the adaptive replacement cache balances recency and frequency; ignores one-time sequential requests and low temporal locality; outperform LRU on most traces and cache sizes [»megiN4_2004]
Subtopic: problems with working sets
Quote: reuse distance is the number of distinct data elements between accesses of the same data element; also called LRU stack distance [»shenX1_2007] Quote: approximate program locality with easily-obtained time distance histograms; 17x faster, 99% accurate for cache reuse, 94% for element reuse
Quote: if organize data for predictable working set size, get unpredictable response times in interactive systems; need random storage access
[»bereG9_1982]
Topic: data caching (35 items)
Topic: memory cache (29 items)
Topic: memory management by age (18 items)
Topic: memory management by paging (23 items)
Topic: memory management for programs and modules (12 items)
Topic: virtual memory
(32 items)