18
A New Prefetch Cache Scheme Shun-Zheng Yu; Kobayashi, H. Department of Electrical Engineering E-Quad, Princeton University. Proceedings of IEEE Conference on Global Telecommunications , 2000. GLOBECOM '00. IEEE , Volume: 1 , 2000 Page(s): 350 - 355

A New Prefetch Cache Scheme Shun-Zheng Yu; Kobayashi, H. Department of Electrical Engineering E-Quad, Princeton University. Proceedings of IEEE Conference

  • View
    228

  • Download
    1

Embed Size (px)

Citation preview

A New Prefetch Cache Scheme

Shun-Zheng Yu; Kobayashi, H. Department of Electrical Engineering

E-Quad, Princeton University.Proceedings of IEEE Conference on Global

Telecommunications , 2000. GLOBECOM '00. IEEE , Volume: 1 , 2000 Page(s): 350 -355

   

Outline

1.Introduction 2.Some Preparation 3.A New Prefetch Scheme 4.Numerical Result

1.Instruction

Conventional:

--probability of each document

--popularity of the document New:

--average latency

--hit probability

--cache capacity

--required bandwidth

Instruction(cont.)

Classification of prefetch: --predictive prefetching(client’s access

history,links between the contents)

--popularity-based prefetching(frequency)

--interactive prefetching(clientproxy)

The goal that we want to achieve --minimizing the latency --maximizing the hit ratio

2.Some preparation HTTP header support

Some preparation(cont.)

Entry_is_fresh=(freshness_lifetime > current_age) Not_Modified=(Validator=Last_Modified time)

Some preparation(cont.)

Some preparation(cont.)

Some preparation(cont.)

(A) min latency

Some preparation(cont.)

(B) max hit ratio

Some preparation(cont.)

(C) min require bandwidth

3.A New Prefetch Scheme

Four Constraints

--Increased cache capacity ΔC:

--Increased bandwidth usage ΔB:

A New Prefetch Scheme

--Tolerable latency Lo:

--Minimum hit probability required Po:

A New Prefetch Scheme

How to operate this scheme 1.Determine the Rn,R,μn,Sn,ΔTn 2.To calculate 3.Sort in the descending order 4.Determine the number r of prefetching

documents(by 4 constraints) 5.Choose the largest r documents , and

prefetch them. 6.Repeat the procedure from 1 to 5 with a given

statistical cycle(hours,days)

4.Numerical Result

Numerical Result(cont.)

Numerical Result(cont.)

Numerical Result(cont.)