Network vs. - Les défis de la nouvelle génération

301 views

Published on

Published in: Technology
0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total views
301
On SlideShare
0
From Embeds
0
Number of Embeds
1
Actions
Shares
0
Downloads
2
Comments
0
Likes
0
Embeds 0
No embeds

No notes for slide

Network vs. - Les défis de la nouvelle génération

  1. 1. Network vs. Server Issues in end-to-end performance Christian Huitema [email_address] June 18, 2000
  2. 2. The Internet grows, and grows… Number of hosts (millions) in the Internet (Netsizer by Telcordia)
  3. 3. Network versus servers: who is trailing whom? <ul><li>Network vs. server performance, today </li></ul><ul><li>State of the network, </li></ul><ul><li>The special case of the DNS, </li></ul><ul><li>The 10 Gigabit challenge, </li></ul><ul><li>Going end-to-end? </li></ul>
  4. 4. Network vs. server performance, today
  5. 5. Measuring the components of the web delay get address connect accept prepare transmit HTTP GET receive close server network delay Get address only once (cache). Connect only once (HTTP 1.1), Pipeline. DNS DNS
  6. 6. Web performance estimation, 1998-2000 <ul><li>For a given client/server pair: </li></ul><ul><ul><li>Measure the address resolution delay, connection delay, delay to first packet, delay to transmission of page </li></ul></ul><ul><li>Repeat every day: </li></ul><ul><ul><li>For a fixed set of 100 “large” servers, </li></ul></ul><ul><ul><li>For a set of 100 “random” servers, provided from randomized requests to Altavista, Lycos, Google, etc. </li></ul></ul>
  7. 7. Observed bandwidth seems to grow over time
  8. 8. The servers seem to contribute 40% of the delay
  9. 9. Feedback Loop: Why Capacity Doubles Every 8 Months Providers increase the network capacity More users, more demand, faster modems More contents, more servers, faster feeds
  10. 10. Network vs. server: what is the state of the network?
  11. 11. First network characteristic: Packet loss rates
  12. 12. Second characteristic: Round trip time
  13. 13. Taking a closer look at the round trip time distribution
  14. 14. Likely evolution of the network <ul><li>Accommodate growing demand: </li></ul><ul><ul><li>Cable, DSL, other broadband </li></ul></ul><ul><ul><li>Mobile phones, appliances </li></ul></ul><ul><li>Likely evolution: </li></ul><ul><ul><li>Average bandwidth grows -> 100 kbps </li></ul></ul><ul><ul><li>Median delays -> distance/speed of light </li></ul></ul><ul><ul><li>Delay deviation diminishes, as predicted by Paul Kenny’s work. </li></ul></ul>
  15. 15. Network vs. server: The special case of the DNS
  16. 16. DNS: an amplifier of the Internet Performance Local Server Root, .Com Example.Com Many network traversals, many servers.
  17. 17. Resolution Delay, Random DNS Names
  18. 18. DNS delays don’t seem to improve at Internet Speed… Fraction of DNS resolution delays over 2 seconds
  19. 19. A clear case of “server vs. network” <ul><li>A.ROOT-SERVERS.NET loss: 33.9% </li></ul><ul><li>B.ROOT-SERVERS.NET loss: 3.4% </li></ul><ul><li>C.ROOT-SERVERS.NET loss: 21.1% </li></ul><ul><li>D.ROOT-SERVERS.NET loss: 2.6% </li></ul><ul><li>E.ROOT-SERVERS.NET loss: 12.3% </li></ul><ul><li>F.GTLD-SERVERS.NET loss: 4.0% </li></ul><ul><li>F.ROOT-SERVERS.NET loss: 12.3% </li></ul><ul><li>G.ROOT-SERVERS.NET loss: 62.1% </li></ul><ul><li>H.ROOT-SERVERS.NET loss: 20.9% </li></ul><ul><li>I.ROOT-SERVERS.NET loss: 7.1% </li></ul><ul><li>J.GTLD-SERVERS.NET loss: 8.3% </li></ul><ul><li>K.GTLD-SERVERS.NET loss: 2.2% </li></ul>Limits of these data: One single test, One single day!
  20. 20. How will the DNS situation evolve? <ul><li>No clear virtuous cycle: </li></ul><ul><ul><li>Registration of new names is a for profit venture (NSI, and wannabes) </li></ul></ul><ul><ul><li>Maintenance of root is a non profit service with political constraints (ICANN) </li></ul></ul><ul><li>Solutions from servers, software: </li></ul><ul><ul><li>Less reliance on root? </li></ul></ul><ul><ul><li>More reliance on caches (load balancing?) </li></ul></ul><ul><ul><li>Less reliance on the DNS! </li></ul></ul>
  21. 21. Network vs. server: The 10 Gigabit challenge
  22. 22. Servers are getting better but is this enough? <ul><li>Internet2 Land Speed Record award: </li></ul><ul><ul><li>&quot;We hope this competition gets people thinking about enabling really revolutionary Internet applications,&quot; - Jim Gray </li></ul></ul><ul><ul><li>&quot;To realize Internet's full potential, end-to-end network performance needs to take a huge leap forward,&quot; - Gordon Bell </li></ul></ul><ul><li>The records: </li></ul><ul><ul><li>751.362 Mbps over 5,626 km (single TCP) </li></ul></ul><ul><ul><li>957.369 Mbps over 5,626 km (2 TCP) </li></ul></ul>
  23. 23. The 10 Gigabit challenge, or the limits of Moore’s law
  24. 24. Meeting the challenge: part 1, ever better servers <ul><li>Get stronger: </li></ul><ul><ul><li>Faster CPU, better memory architecture </li></ul></ul><ul><ul><li>Faster I/O </li></ul></ul><ul><li>Get smarter: </li></ul><ul><ul><li>HTTP 1.1, TCP, IPv6… </li></ul></ul><ul><ul><li>Better protocol design(s) </li></ul></ul><ul><ul><li>Better software (doing that) </li></ul></ul><ul><li>Get help: </li></ul><ul><ul><li>Offload packet processing (done that) </li></ul></ul><ul><ul><li>Offload encryption (done for IPSEC) </li></ul></ul><ul><ul><li>Offload key management (uh?) </li></ul></ul>
  25. 25. Meeting the challenge: part 2, larger clusters <ul><li>Servers x2 every 18 month. </li></ul><ul><li>Can we manage a cluster of 100,000 servers? </li></ul><ul><ul><li>Load balancing, Software upgrade, Failover </li></ul></ul><ul><li>Can we develop distributed applications? </li></ul>
  26. 26. Meeting the challenge: part 3, distribution networks <ul><li>Push the “content” near the “consumers” </li></ul><ul><li>Is this needed in a 10 Gigabit world? </li></ul><ul><li>Does it work for more than GIF files? </li></ul>
  27. 27. Evolution of server, server farms and distribution <ul><li>Handling variability </li></ul><ul><ul><li>Heavy tail distribution of demand means flash crowds, etc. </li></ul></ul><ul><ul><li>Could incite to a “market of servers”, or to “adaptive servers.” </li></ul></ul><ul><li>Pushing up hill </li></ul><ul><ul><li>Need a serious business case for investment in new servers </li></ul></ul><ul><ul><li>Constantly testing the limits of the technology </li></ul></ul>
  28. 28. Network vs. server: Going end-to-end?
  29. 29. End to end distribution: NAPSTER, Gnutella Rendezvous, directory
  30. 30. End to end communication: real time audio, video, games SIP proxy SIP proxy
  31. 31. Meeting the challenge: distribute the load! <ul><li>The return of the end-to-end argument </li></ul><ul><ul><li>Each new client brings its own resource </li></ul></ul><ul><ul><li>The only known way of scaling </li></ul></ul><ul><li>Only place in the servers the functions that cannot possibly placed in the client! </li></ul>
  32. 32. Network vs. server: Thank You!

×