Titlepage Improving Content Delivery Using Provider-aided Distance Information Ingmar Poese Benjamin Frank Bernhard Ager Georgios Smaragdakis Anja Feldmann { ingmar|bfrank|bernhard|georgios|anja } @net.t-labs.tu-berlin.de Technische Universtit¨ at Berlin/Deutsche Telekom Laboratories November 1 2010 Poese et al (TUB/DT) Improving Content Delivery Using Provider-aided Distance Information Nov 1 2010 1
Outline 1 Background 2 Diversity in cache locations 3 Leveraging diversity with PaDIS 4 Evaluation Poese et al (TUB/DT) Improving Content Delivery Using Provider-aided Distance Information Nov 1 2010 2
Background Content is King World-wide World-wide World-wide World-wide Sandvine Sandvine Germany Germany Germany Germany MidEast MidEast ipoque ipoque ipoque ipoque Maier Maier Arbor Arbor 2008 2009 2007 2009 2007 2009 2007 2009 2008 2009 100% P2P more likely* 80% Web at 18% Streaming Usenet (NNTP) BitT orrent/P2P 60% eDonkey Other/known Unclassified 40% *Based on limited dataset using payload inspection 20% 0% Fabian Schneider: Analysis of New Trends in the Web from a Network Perspective, 2010 • Web and streaming are dominating Internet traffic • both run over HTTP • Contribute up to 60 % of the volume Poese et al (TUB/DT) Improving Content Delivery Using Provider-aided Distance Information Nov 1 2010 3
Background CDN deployment C C C C C C C Client C C C C C C C C C C C Poese et al (TUB/DT) Improving Content Delivery Using Provider-aided Distance Information Nov 1 2010 4
Background CDN deployment C C C C C C C Client C C C C C C C C C Caches C C • Inside ISPs • By subnet • By location • Assignment via DNS • By DNS redirection Poese et al (TUB/DT) Improving Content Delivery Using Provider-aided Distance Information Nov 1 2010 4
Background CDN cache selection DNS based cache selection auth. DNS C 1 Client queries resolver 2 4 Client 2 Redirect to CDN Internet 3 CDN chooses cache(s) 1 3 resolver • Return via resolver CDN DNS 4 Connect to cache Poese et al (TUB/DT) Improving Content Delivery Using Provider-aided Distance Information Nov 1 2010 5
Background CDN cache selection DNS based cache selection auth. DNS C 1 Client queries resolver 2 4 Client 2 Redirect to CDN Internet 3 CDN chooses cache(s) 1 3 resolver • Return via resolver CDN DNS 4 Connect to cache Known metrics Unknown metrics • Cache load • Exact position • Content availability • Path properties Poese et al (TUB/DT) Improving Content Delivery Using Provider-aided Distance Information Nov 1 2010 5
Outline 1 Background 2 Diversity in cache locations 3 Leveraging diversity with PaDIS 4 Evaluation Poese et al (TUB/DT) Improving Content Delivery Using Provider-aided Distance Information Nov 1 2010 6
Diversity in cache locations Data Other Customers of AS AS Edge Router Customers Downstream of Monitoring Point Monitoring Point Internet/other ASes Broadband Access Router Click stream data • Anonymized trace from a POP in a large European ISP. • Trace spans over 20.000 customers and 14 days • total of 1.2 billion requests (89 million/day) • Examine top 10,000 hostnames • Exposed location diversity • Potential for content delivery Poese et al (TUB/DT) Improving Content Delivery Using Provider-aided Distance Information Nov 1 2010 7
Diversity in cache locations Location diversity per hostname 1 0.1 0.01 CCDF 0.001 0.0001 1e-05 1 2 3 4 5 6 7 8 Avg. Number Locations per Hostname Poese et al (TUB/DT) Improving Content Delivery Using Provider-aided Distance Information Nov 1 2010 8
Diversity in cache locations Location diversity per content provider 1 0.9 0.8 0.7 0.6 CDF 0.5 Ubiquitous Caches 0.4 • Serve anything 0.3 • Fetch missing 0.2 0.1 0 1 10 100 # of Different Locations Poese et al (TUB/DT) Improving Content Delivery Using Provider-aided Distance Information Nov 1 2010 9
Leveraging diversity with PaDIS Opportunities for ISPs Current situation • CDNs do not expose location diversity • Cache selection without accurate client position • Unknown path properties to content consumer ⇒ Can content delivery be improved with location diversity ? Poese et al (TUB/DT) Improving Content Delivery Using Provider-aided Distance Information Nov 1 2010 10
Leveraging diversity with PaDIS Opportunities for ISPs Current situation • CDNs do not expose location diversity • Cache selection without accurate client position • Unknown path properties to content consumer ⇒ Can content delivery be improved with location diversity ? P rovider- a ided D istance I nformation S ystem ( PaDIS ) • PaDIS can utilize the diversity in paths to locations • Improves application performance and Quality of experience • Reduce page load delay • Reduce download time for large files • Removes the need for in-accurate active measurement • ISPs gain influence on the path selection to locations Poese et al (TUB/DT) Improving Content Delivery Using Provider-aided Distance Information Nov 1 2010 10
Leveraging diversity with PaDIS Status Quo Steps 1 DNS Query 2 Find auth. DNS server 3 Receive auth. DNS answer 4 DNS resolver forwards reply External DNS 3 2 ISP DNS Resolver 1 4 Client Poese et al (TUB/DT) Improving Content Delivery Using Provider-aided Distance Information Nov 1 2010 11
Leveraging diversity with PaDIS PaDIS usage example Steps 1 DNS Query 2 Find auth. DNS server 3 Receive auth. DNS answer 4 Send Answer to PaDIS External DNS 5 PaDIS aggregates and reorders known IPs 3 6 DNS resolver sends top ranked IPs 2 back to client ISP DNS Resolver 5 1 4 PaDIS 6 Client Poese et al (TUB/DT) Improving Content Delivery Using Provider-aided Distance Information Nov 1 2010 11
Leveraging diversity with PaDIS PaDIS usage example PaDIS Properties • Operated by the ISP • Up-to-date network information • Knowledge of CDN caches • No architecture change needed • Transparent to consumer and CDN Poese et al (TUB/DT) Improving Content Delivery Using Provider-aided Distance Information Nov 1 2010 12
Evaluation Experiment setup for CDN C C C C C C C resolver 1 C C C C C C C C C Client 2 C C Poese et al (TUB/DT) Improving Content Delivery Using Provider-aided Distance Information Nov 1 2010 13
Evaluation Experiment setup for CDN C C C PaDIS PaDIS C C 2 C C resolver 1 C C C C C C C C C Client 3 C C Poese et al (TUB/DT) Improving Content Delivery Using Provider-aided Distance Information Nov 1 2010 14
Evaluation Experiment setup for CDN C C C PaDIS PaDIS C C 2 C C resolver 1 C C C C C C C C C Client 3 C C Statistics Algorithm • 124 locations • Download file from all locations • 11 files • Compare CDN selection with PaDIS Poese et al (TUB/DT) Improving Content Delivery Using Provider-aided Distance Information Nov 1 2010 14
Evaluation Selected result for CDN Downloading a 510K file from 124 locations every hour 510K 3.0 recommended PaDIS download time in s 2.5 2.0 1.5 1.0 0.5 00:00 12:00 24:00 time Poese et al (TUB/DT) Improving Content Delivery Using Provider-aided Distance Information Nov 1 2010 15
Evaluation Expanding the scope Using PaDIS on a One-Click-Hoster (OCH) ISP-1 ISP Peering P e OCH Link e r i n g L i n k Peering OCH File ISP-2 Peering Link Link ISP-3 Repeat experiment setup with an OCH • Select peering link instead of cache location • Repeatedly download a (60 Mbyte) File via all peerings • Evaluate link selection • Compare download times Poese et al (TUB/DT) Improving Content Delivery Using Provider-aided Distance Information Nov 1 2010 16
Evaluation Download time Evaluation Downloading a 60MByte file every two hours via all peering links direct Peering via ISP-2 via ISP-1 via ISP-3 Download Time (in Seconds) 200 100 50 20 Thu Fri Sat Sun Mon Tue Wed Thu Time Poese et al (TUB/DT) Improving Content Delivery Using Provider-aided Distance Information Nov 1 2010 17
Evaluation Summary Summary • CDNs do not expose location diversity consumers • PaDIS can expose and utilize the diversity • Localize traffic • Decrease delay and download times • Give power back to the ISP • The experiments (CDN + OCH) show a signifcant reduction in download time Future work • Quantify traffic reduction for ISPs when using PaDIS • Can PaDIS be used for traffic engineering by ISPs ? Poese et al (TUB/DT) Improving Content Delivery Using Provider-aided Distance Information Nov 1 2010 18
Recommend
More recommend