SLIDE 1
2
Overview
■ Problem: DNS overhead ■ Solution: Web-DNS Cooperation ■ Experiment: Squid log analysis ■ Conclusions ■ Future Work
Cross-Domain Cooperation for Small Clients Amy S. Hughes Joe Touch - - PowerPoint PPT Presentation
Cross-Domain Cooperation for Small Clients Amy S. Hughes Joe Touch - - PowerPoint PPT Presentation
Cross-Domain Cooperation for Small Clients Amy S. Hughes Joe Touch USC/ISI Overview Problem: DNS overhead Solution: Web-DNS Cooperation Experiment: Squid log analysis Conclusions Future Work 2 DNS Overhead in Web
SLIDE 2
SLIDE 3
3
DNS Overhead in Web Transactions
■ DNS request is a large part of web
transaction
■ DNS request dominates as: – Bandwidth increases – Persistent connections reduce overhead – Latency increases ■ DNS is multiple RTTs
A = Connection Goodput B = Connection Establishment C = DNS Requests A + B + C = Total Connection Time
A + B + C A + B + C C C <
SLIDE 4
4
Web Connection Components
Client Server
Request Response SYN SYN/ACK ACK GET RESPONSE FIN FIN/ACK ACK Time
DNS Server
4 5 1 2 3 1: DNS 2: Connect 3: First-Response 4: Start-total 5: End-total
SLIDE 5
5
DNS overhead with low latency
LAN/remote requests LAN/local requests
Server DNS Client
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0.001 0.01 0.1 1 10 100 Fraction of R eq uests Seconds (log) DNS Connect First-response End-total 1 3 2 5
Server DNS Client
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0.001 0.01 0.1 1 10 100 Fractio n of Requests Seconds (log) DNS Connect First-response End-total 1 2 3 5
SLIDE 6
6
DNS overhead with high latency
ISDN/remote requests ISDN/local requests
Server DNS Client Server DNS Client
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0.001 0.01 0.1 1 10 100 Fractio n of Requests Seconds (log) DNS Connect First-response End-total 2 3 5 1 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0.001 0.01 0.1 1 10 100 Fractio n of Requests Seconds (log) DNS Connect First-response End-total 1 2 3 5
SLIDE 7
7
DNS Reuse and Costs
■ Squid logs: 10-15% DNS misses ■ 2MB cache upper bound
SLIDE 8
8
Cache Anticipation
■ Web Cache
– Request stream related to item content – Anticipation possible
■ DNS Cache
– No item relation to request stream – No anticipation opportunity
SLIDE 9
9
Web-DNS Cooperation
■ Opportunity
– Web request requires DNS information – Cooperation possible
■ Solution
– DNS cache on local client – Web lookahead to anticipate DNS requests
SLIDE 10
10
DNS Anticipated Cache Size
DNS Cache Size / Time
5000 10000 15000 20000 25000 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 S ize of Cache ( # o f ent ries) Time (hours) Simple Caching Anticipation
SLIDE 11
11
DNS Misses and Reduction
10 20 30 40 50 60 70 80 90 100 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Actu al Miss Rate Time (hours) Cache Miss Rate Anticipation Miss Rate 10 20 30 40 50 60 70 80 90 100 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Miss Rate C hange Time (hours) Anticipation Benefit
DNS Miss Rate Miss Rate Change
SLIDE 12
12
DNS Miss Reduction
Magnification of MISS Rate Change
5 10 15 20 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Miss Rate Ch ange Time (hours) Anticipation Benefit
SLIDE 13
13
Prior and Related Work
■ Web Log Analysis ■ Web Anticipation ■ Web Cooperation
– Squid – LSAM – Adaptive Web Caching
SLIDE 14
14
Conclusions
■ DNS caches must be local on client
machines to be useful
– 90% benefit – esp on ISDN connections
■ DNS-Web cooperation needs more
exploration
– 15% reduced misses – 3x space increase (<6MB total)
SLIDE 15
15
Future Work
■ Analyze real-time client traces – Squid logs wrong place in cache hierarchy – Real-time tracing allows examination of time components ■ Define DNS hits and misses – Some DNS misses are partial hits due to mulitple RTTs ■ Implementation of Cross-Domain system – Measure real benefits – Examine DNS aggregation