Aging Through Cascaded Caches: Performance Issues in the Distribution of Web Content'

1
Aging Through Cascaded CachesPerformance Issues
in the Distribution of Web Content.

• Edith Cohen
• ATT Labs-research

Haim Kaplan Tel-Aviv University
2
HTTP Freshness Control
Cache-directives

• Cached copies have
• Freshness lifetime
• Age (elapsed time since fetched from origin)
• TTL (Time to Live)
• freshness lifetime age
• Expired copies must be validated before they can
  be used (request constitutes a cache miss).

header
Body (content)
3
Aging of Copies
Freshness Lifetime 10 hours
4
Aging of Copies
900am
Age 1 TTL 9
Freshness Lifetime 10 hours
5
Aging of Copies
600pm
Freshness Lifetime 10 hours
6
Aging thru Cascaded Caches
proxy caches
reverse-proxy cache
7
Aging thru Cascaded Caches
proxy caches
500pm
reverse-proxy cache
Age 9 TTL 1
8
Aging thru Cascaded Caches
proxy caches
600pm
reverse-proxy cache
!! !!
9
Aging thru Cascaded Caches
proxy caches
600pm
reverse-proxy cache
10
TTL of a Cached Copy
TTL
From Origin
M
M
From Cache
Freshness-lifetime
Requests at client cache
t
11
Age-Induced Performance Issues for Cascaded Caches

• Caches are often cascaded (path between web
  server and end-user includes 2 or more caches.).
• Copies obtained thru a cache are less effective
  than copies obtained thru an origin server.
• Reverse proxies increase validation traffic
  !!
• More misses at downstream caches mean
• Increased traffic between cascaded caches.
• Increased user-perceived latency.

12
Research Questions

• How does miss-rate depend on the configuration of
  upstream cache(s) and on request patterns ?
• Can upstream caches improve performance by
  proactively reducing content age ? how?
• Can downstream caches improve performance by
  better selection or use of a source?

Our analysis

• Request sequences Arbitrary, Poisson, Pareto,
  fixed-frequency, Traces.
• Models for Cache/Source/Object relation
  Authoritative, Independent, Exclusive.

13
Basic Relationship Modelscache/source/object

• Authoritative Origin server 0 age copies.
• Exclusive all misses directed to the same
  cache.
• Independent each miss is directed to a different
  independent upstream cache.

14
Basic Models
Object has fixed freshness-lifetime of T. Miss at
time t results in a copy with age

• Authoritative age(t) 0
• Exclusive age(t) T - (ta) mod T
• Independent age(t) e U0,T

Theorem On all sequences, the number of misses
obeys Authoritative lt Exclusive lt Independent
Theorem Exclusive lt 2Authoritative Independen
t lt eAuthoritative
15
TTL of Supplied Copy
Source
TTL
Freshness-lifetime
t
Requests Received at source
16
How Much More Traffic?
Miss-rate for different configurations
17
Rejuvenation at Source Caches
Rejuvenation refresh your copy pre-term once its
TTL drops below a certain fraction v of the
Lifetime duration.
TTL
v0.5
source
24h
12h
t
Requests at client
18
Rejuvenations Basic Tradeoff
Downstream Client caches
Upstream cache
origin
Is increase/decrease monotone in V (?)
19
Interesting Dependence on V

• Independent(v) ltgt Exclusive(v)
• Independent(v) is monotone if v1 gt v2,
• Independent(v1) gt Independent(v2)
• Exclusive(v) is not monotone
• (miss-rate can increase !!)
• Integral 1/v (synchronized rejuvenation)
  Exclusive(v) lt Independent(v) and is monotone
  (Pareto, Poisson, not with fixed-frequency).

20
(No Transcript)
21
(No Transcript)
22
How Can Non-integral 1/v Increase Client Misses?
Copy at client is not synchronized with
source. When it expires, the rejuv source has an
aged copy.
TTL
Upstream Cache
Downstream Client Cache
Pre-term refreshes
Freshness-lifetime
t
23
Why Integral 1/v Works Well?
Cached copies remain synchronized
TTL
Upstream Cache
v0.5
Downstream Client Cache
Pre-term refreshes
Freshness-lifetime
t
24
Some Conclusions

• Configuration Origin (Authoritative) is best.
  Otherwise, use a consistent upstream cache per
  object (Exclusive).
• No-cache request headers resulting sporadic
  refreshes may increase misses at other client
  caches. (But it is possible to compensate).
• Rejuvenation potentially very effective, but a
  good parameter setting (synchronized refreshes)
  is crucial.
• Behavior patterns Similar for Poisson, Pareto,
  traces, (temporal locality). Different for
  fixed-frequency.
• For more go to http//www.research.att.com/edith
  
• Full versions of Cohen, Kaplan SIGCOMM 2001
• Cohen, Halperin,
  Kaplan, ICALP 2001