AMP: ProgramContext Specific Buffer Caching

1
AMP Program-Context Specific Buffer Caching

• Feng Zhou, Rob von Behren, Eric Brewer
• University of California, Berkeley
• Usenix tech conf 2005, April 14, 2005

2
Buffer caching beyond LRU

• LRU good for workload with temporal locality, but
  very bad for loops and scans
• Large (gtcache size) one-time scan evicts all
  blocks
• Large (gtcache size) loops get zero hits
• Renewed interests in buffer caching
• Adaptation ARC (FAST03), CAR (FAST04)
• Detection UBM (OSDI00), DEAR (Usenix99), PCC
  (OSDI04)
• basic idea detect loops and scans and apply MRU
  or variant

3
Adaptive Multi-Policy (AMP) caching

• Robust algorithm for detecting looping patterns
• more robust against local reorders or small
  spurious loops
• Randomized scheme for managing many cache
  partitions
• much cheaper than previous precise methods
• PC-specific caching that classifies disk accesses
  by program contexts
• independent from PCC by Gniady et al (OSDI04)

4
PC-specific Buffer Caching

• Program context the current PC all return
  addresses on the call stack

Ideal policies 1 MRU 2 LFU
3 LRU
5
Loop detection scheme

• Intuition measure the average recency of the
  blocks accessed
• For the i-th access
• Li list of all previously accessed blocks,
  ordered from the oldest to the most recent by
  their last access time.
• Li length of Li
• pi position in Li of the block accessed (0 to
  Li-1)
• Define the recency of the access as,

6
Loop detection scheme cont.

• Average recency R of the whole sequence is the
  average of all defined Ri (0ltRlt1)
• Detection result
• loop, if R lt Tloop (e.g. 0.4)
• temporally clustered, if R gt Ttc (e.g. 0.6)
• others, o.w. (near 0.5)

7
Loop detection example 1

• Block sequence 1 2 3 1 2 3

• R 0, detected pattern is loop

8
Loop detection example 2

• Block sequence 1 2 3 4 4 3 4 5 6 5 6, R 0.79

9
Randomized Cache Partition Management

• Need to decide cache sizes devoted to each PC
• Marginal gain (MG) the expected number of extra
  hits over unit time if one extra block is
  allocate
• Local optimum when every partition has the same
  MG
• Concrete scheme
• Expand the LRU partition by one if ghost buffer
  hit
• Expand an MRU partition by one every
  loop_size/ghost_buffer_size accesses to the
  partition
• Compared to previous schemes
• No need to find partition smallest MG (O(logN))

10
Detection of synthesized sequences
tctemporally clustered Colored detection
results are wrongClassifying tc as other is
deemed correct.
11
Simulation dbt3 (tpc-h)
12
Simulation scan
13
Implementation

• Kernel patch for Linux 2.6
• Shortens time to index Linux source code using
  glimpseindex by up to 13 (read traffic down 43)
• Shortens time to complete DBT3 (tpc-h) DB
  workload by 9.6 (read traffic down 24)
• Linux implementation and simulatorhttp//www.cs.
  berkeley.edu/zf/amp

14
(No Transcript)
15
DBT3 on AMP implementation

• Overall execution time reduced by 9.6 (1091 secs
  ?986 secs)
• Disk reads reduced by 24.8 (15.4GB ? 11.6GB),
  writes 6.5

16
Correctness of partition size adaptation

• During time period t, number of expansion of ARC
  is
• 
  (B1B2ghost_buffer_size)
• The number of expansions of an MRU partition i is
• They are both proportional to their respective MG
  with the same constant