improving www performance using domain-top approach to prefetching seung won shin, byeong hag seong...
Post on 21-Dec-2015
216 views
TRANSCRIPT
Improving WWW Performance Using Domain-Top approach to
prefetching
Seung Won Shin, Byeong Hag Seong and Daeyeon park
Conference/Exhibition on High Performance Computing in the Asia-Pacific Region, 2000.
Proceedings. The Fourth International , Volume: 2 , 2000 Page(s): 738 -746 vol.2
Mike Tien
1.Introduction Top-10 --predicts the next requests by the prediction table consisting of
the documents that accessed very frequently. --ad: It is simple and imposes very small burden on the proxy. --disa: It does only collect documents access frequently without analyzing the characteristics of requests.It does not cope with the rapid change of clients.
Access-Tree --prediction table made by tree structure for measuring of correlation. --ad: It can predict the future request more accurately by probability. --disa: It has to manage complex data structure prediction table. It updated almost every time for each client request and determines the access probability of each resource.
Introduction(cont.)
Domain-Top When most of clients find a domain that they want,
they wait and watch the several documents in the domain.
Top-domain--The domains that are requested frequently
Top-Documents—the documents that are demanded most frequently in each Top-domain.
Rank-list—It is a simple list, and it has an enough information to predict client future requests.
2.Prefetching Clients----proxies----servers There are two types of user-level processes.
-- http-daemon—with support for persistent connections and
pipeline requests.
-- prediction-daemon—only manages and analyzes access log file in the proxy,not directly with client request.
Prefetching(cont.)
About Rank List -- How many domains are selected and how many documents
are selected for each domain? -- a client moves about 20 domains in a day mainly.(by analyzed) -- prefetch 8 resources at once.
( the number of documents for prefetching increases, hit ratio of the prefetching grows, but its growing rate is so small and saturated in a moment )
Prefetching(cont.)
Persistent connection and Pipelining Request through GETLIST method
-- We use HTTP 1.1 persistent connection for retrieving the
multiple resources.This makes it possible to take several
resources in one connection.
-- GETLIST allows a client to request a set documents or images
from a server.( a series of GETs)
3.Simulation Result and Performance
Average size of all files that were prefetched is 40 KB.
No more than 8 resources can be prefetched during one prefetching time period.
4.Summary
Prediction by Domain-Top approach -- Top-10 -- Access-Tree -- Domain-Top Lightweight prediction Table: Rank List Small variation of hit ratio Small change of the proxy Small burden on the network -- persistent connection -- GETLIST