Hide Notes 

Slide 1: CS 7270 Networked Applications & Services Lecture-7 Web performance Part-1: Content Distribution Nets 1
Slide 2: How Akamai Works (from Srini Seshan s CMU lecture) cnn.com (content provider) Get foo.jpg Get index.h tml 1 11 12 DNS root server Akamai server 2 3 4 5 6 7 8 9 Akamai high-level DNS server Akamai low-level DNS server Nearby matching Akamai server End-user 10 Get /cnn.com/foo.jpg 2
Slide 3: Akamai Subsequent Requests DNS root server Akamai server cnn.com (content provider) Get index. html 1 2 7 8 9 Akamai high-level DNS server Akamai low-level DNS server Nearby matching Akamai server End-user 10 Get /cnn.com/foo.jp g 3
Slide 4: Reading  On the Use and Performance of Content Dist by B. Krishnamurthy et al.   Appeared in IMC 01 Highly influential paper (but a bit outdated now) 4
Slide 5: On the Use and Performance of Content Distribution Networks Yin Zhang Joint work with Balachander Krishnamurthy and Craig Wills AT&T Labs Research, WPI {yzhang,bala}@research.att.com, cew@cs.wpi.edu ACM SIGCOMM Internet Measurement Workshop November, 2001
Slide 6: Motivation  What is a CDN?  A network of servers delivering content on behalf of an origin site A number of CDN companies   State of CDNs  E.g. Akamai, Digital Island, Speedera E.g., CNN, CNBC, &  Used by many popular origin sites   Little has been published on the use and performance of existing CDNs 6
Slide 7: Research Questions to Answer      What CDN techniques are being used? What is the extent to which CDNs are being used by popular origin sites? What is the nature of CDN-served content? What methodology can be used to measure the relative performance of CDNs? How are specific CDNs performing both relative to origin servers and among themselves? This talk tries to answer them based on a large-scale, client-centric study conducted in Sept. 2000 and Jan. 2001 7
Slide 8: What CDN redirection techniques are being used? CDN Server Request/ Response CDN Name Server  Techniques examined  DNS redirection (DR)   Full-site delivery (DR-F) Partial-site delivery (DR-P)  CDN server IP   URL rewriting (UR) Hybrid scheme (URDR)  URL rewriting + DNS redirection Techniques NOT examined     Client CDN server name Origin Server Manual hyperlink selection HTTP redirection Layer 4 switching Layer 7 switching 8
Slide 9: How widely are CDNs being used?  Sources of data Type Periodic crawl Proxy log Datasets HotMM127 URL588-MM500 LMC NLANR Date/Duration 2 months: Nov. & Dec. 2000 1 week in Sept. 2000 1 week in Jan. 2001 Sites 127 1030 3 9  CDN use by popular sites Nov. 1999 Dec. 2000 1-2% out of ~600 [KW00] HotMM127: 31% URL588-MM500: 17% (Akamai: 98%) (Akamai: 85%) 9
Slide 10: Nature of CDN-served Content  Daily change characteristics of CDN-served objects Dataset #Objects Prev. seen URL Prev. seen URL w/ changes HotMM127 24.9K 89% 2.2% URL588-MM500 75.0K 86% 3.2%  Nature of HTTP-requested CDN content    Images account for 96-98% CDN-served objects, or 40-60% CDN-served bytes Akamai serves 85-98% CDN-served objects (bytes) Cache hit rates of CDN-served images are generally 20-30% higher than non-CDN served images 10
Slide 11: Performance Study: Methodology CDN Server CDN Name Server 1 Get CDN server IP address 1 URL rewriting first get CDN server name 2 3 2 1 Warm up CDN cache 3 Retrieve pages using httperf Parallel-1.0 4 HTTP/1.0 Serial-1.1 -- 2 persistent HTTP/1.1 Pipeline-1.1 1 pipelined HTTP/1.1 1 Client Origin Server General Methodology: From N client sites periodically download pages from different CDNs and origin sites. 11
Slide 12: Content for Performance Study  Challenge:  Different CDNs have different customers. How to compare apples to apples ?  Solution: Canonical Pages  Create template page based on distributions of the number and size of embedded images at popular sites  In our study, we download 54 images and record download time for the first 6, 12, 18, 54 images.  For each CDN, construct a canonical page with a list of image URLs currently served by the CDN from a single origin site, that closely match the sizes in the template page. 12
Slide 13: Measurement Infrastructure  CDNs Technique CDNs DR-F Adero DR-P Akamai, Speedera, Digital Island UR Clearway URDR Fasttide *AT&T ICDS NOT tested due to conflict of interest.  Origin sites   US: Amazon, Bloomberg, CNN, ESPN, MTV, NASA, Playboy, Sony, Yahoo International: 2 Europe, 2 Asia, 1 South America, 1 Australia 24 NIMI client sites in 6 countries    Client sites  NIMI: National Internet Measurement Infrastructure Well-connected: mainly academic and laboratory sites 13
Slide 14: Response Time Results (I) Excluding DNS Lookup Time CDNs generally provide much shorter download time. 14 Cumulative Probability
Slide 15: Response Time Results (II) Including DNS Lookup Time Cumulative Probability DNS overhead is a serious performance bottleneck for some CDNs. 15
Slide 16: Impact of Protocol Options and the Number of Images Mean Download Performance Range for Different Numbers of Images and Protocol Options (Jan. 2001) Protocol Option Parallel-1.0 Serial-1.1 Pipeline-1.1 Site CDN US Origin CDN US Origin CDN US Origin Mean Download Time Range (sec.) 6 images 12 images 18 images 54 images 0.26-0.76 0.40-1.23 0.58-1.53 1.49-3.31 1.63 1.06 2.45 1.46 3.40 1.96 8.42 4.87 0.27-0.53 0.42-0.81 0.61-1.13 1.46-2.52 0.26-0.50 0.37-0.67 0.47-0.88 1.09-2.04 Partial Support CDNs perform significantly better than origin sites, although reducing the number of images (e.g. due to caching) and using HTTP/1.1 options reduces the performance difference. 16
Slide 17: Effectiveness of DNS Load Balancing Small DNS TTLs generally do not improve download times. 17
Slide 18: Effectiveness of DNS Load Balancing (cont d) Parallel-1.0 Download Performance for CDN Server at New and Fixed IP Addresses (Jan. 01) CDN (technique) Adero (DR-F) Akamai (DR-P) Digisle (DR-P) Fasttide (URDR) Speedera (DR-P) Mean completion time (sec.) 5.40 1.15 1.31 2.10 0.72 1.09 1.00 1.21 1.46 0.53 90% completion time (sec.) 9.60 3.05 2.30 4.72 1.53 1.60 3.00 1.70 3.25 1.01 New IP Fixed IP New IP Fixed IP Small DNS TTLs generally do not improve download times in either average or worst case situations. 18
Slide 19: Summary     There is a clear increase in the number and percentage of popular origin sites using CDNs  may have decreased subsequently & CDNs performed significantly better than origin sites, although caching and HTTP/1.1 options both reduce the performance difference Small DNS TTLs generally do not improve client download times in either average or worst case situations Our methodology can be extended to test CDN performance for delivering streaming media  More streaming media results available in the TM version: http://www.research.att.com/~bala/papers/abcd-tm.ps.gz 19