Internet2 Engineering Update.ppt

Document Sample
Internet2 Engineering Update.ppt Powered By Docstoc
					Technology and Administrative
Coordination Issues

Pacific Rim Networking Workshop
Guy Almes    <>

Manoa Valley, Oahu
22 February 2002
            Internet2 Engineering
•   Provide our universities with superlative
    • Performance
    • Functionality
    • Understanding

•   Make superlative networking strategic for
    university research and education
           Technology Issues

•   Multicast

•   IPv6

•   Performance

•   Measurement

•   Security

•   Any Source (Conventional) IPv4 Multicast
    •   Steve Deering's PhD thesis from Stanford
    •   Led to MBONE, then native IP multicast
    •   PIM-Sparse, MBGP, and MSDP

•   Technical Implications
    •   Group g has global significance
    •   Host s creates and joins g and can both send and receive packets
    •   Other hosts can join g and can both send and receive packets
    •   MSDP needed to discover the source(s) sending to g
    •   Each host receives packets from <*,g>
            Issue: Global deployment

•   Careful inclusion of ASM IPv4 Multicast in
    international peering

•   Inclusion of multicast issues on local campuses

•   Bandwidth must be sufficient for all sources to
    all destinations

•   Allocation of group IDs
           Multicast Applications
•   Access Grid and DVTS: distance education
    and conferencing among sets of

•   Streaming Audio/Video

•   Sending files to many destinations, as
    with Digital Fountain
                Scalability and SSM
•   Recall implications of ASM
    •   Global Significance of 'g' value
    •   Any host can join/send to group g

•   SSM being deployed to resolve this
    •   Host s creates a channel <s,g>
    •   Others can subscribe to <s,g>, but only s sends
    •   Source discovery now trivial, so MSDP not needed
    •   g now only has local significance

•   Easy to support in wide area, but new IGMP
•   Applications need to be adapted

•   Clarify motivation for IPv6
    •   End-to-end transparency and global addressability
    •   Supports application innovation, e.g., peer-to-peer

•   Support deployment and engineering expertise
    on networks, especially on campus

•   Anticipate need for first-class support
    •   E.g., 10 Gb/s Abilene upgrade
    •   E.g., Linux, Windows XP
             Issues: Training

•   Within Internet2, IPv6 Training
    • About 8-10 workshops this year
    • First: in Los Angeles, hosted by CENIC, in February
             Issue: Deployment

•   Get some IPv6 on each campus/NRN
    • Tunneled IPv6 over IPv4 works well
    • Performance and network management are limited,
•   Prepare for native peering
    • Abilene will be native IPv6 as part of current upgrade
    • Implications for router selection!

•   Explore applications, DNS, operational
    stability, multicast
               Issue: Performance

•   Tunnels limit performance dramatically
    •   About 30 Mb/s on Cisco 7200, for example

•   Some tunnels will exist for some time

•   But, we must remove tunnels in all
    performance-sensitive paths

•   Thus, remove tunnels from key wide-area
          Issue: Operations

•   IPv6 needs to become a 'normal' protocol

•   Robustness of DNS etc.

•   Mature network management etc.
            End-to-End Performance:
•   In former times, very low bandwidth led to
    (correctly) low expectations
•   Now, serious bandwidth exists
    • TransPac deployment of two OC-12 representative

•   Bandwidth growth will likely continue
    • North America to Europe as a challenging example
             End-to-End Performance:
•   Bandwidth is not the only issue
•   Neither the speed of light nor geographical
    distance across the Pacific have improved!
•   Thus, round-trip times cause problems:
    • Sluggish TCP convergence
    • Interactive applications more difficult

•   Thus, direct physical paths needed
    • Hawaii can play a role here
            End-to-End Performance:
            Packet Loss
•   TCP Throughput 
              MTU / (RTT * PacketLoss)

•   This packet loss include that due to:
    • Congestion
    • Other sources

•   Thus, we need to remove any source of
    non-congestive packet loss
          End-to-End Performance:
•   There is almost always an Ethernet link
    somewhere along a wide-area path, hence
    end-to-end MTU seldom more than 1500

•   But larger MTUs are supported on wide-
    area links, e.g., 9180 on Abilene

•   When performance really matters, work to
    support large end-to-end MTUs
             Threats to
             End to End Performance
•   Fiber problems
    • dirty fiber
    • dim lighting
    • 'not quite right' connectors
             Threats to
             End to End Performance
•   Fiber problems
•   Switches
    • horsepower
    • full vs half-duplex
    • head-of-line blocking
             Threats to
             End to End Performance
•   Fiber problems
•   Switches
•   Inadvertently stingy provisioning
    • mostly communication
    • happens also in international settings
             Threats to
             End to End Performance
•   Fiber problems
•   Switches
•   Inadvertently stingy provisioning
•   Wrong Routing
    • asymmetric
    • best use of Internet2
    • distance
           Threats to
           End to End Performance
•   Fiber problems
•   Switches
•   Inadvertently stingy provisioning
•   Wrong Routing
• Host issues
    • NIC
    • OS / TCP stack
    • CPU
           Perverse Result

•   'Users' think the network is congested or
    that the Internet2 infrastructure cannot
    help them

•   'Planners' think the network is
    underutilized, no further investment
    needed, or that users don't need high
    performance networks

•   Traffic utilization
    •   MRTG, etc., need to be more visible

•   Performance-related measurements
    •   iperf, AMP, Surveyor, etc. along key paths

•   Passive measurements
    •   Netflow becoming mature
    •   OC3MON hardware-based sampling of actual packets
    •   Router support becoming available
             Security: An unusual
             Internet2 Emphasis
•   Aspects of Security
    • Security of the infrastructure
    • Security of user host computers
    • Security of information and privacy

•   In the post-11-Sep environment
    • Society will be less tolerant of lax standards

•   Not a distinctly 'Internet2' concern
    • but one that all our universities share

shenreng9qgrg132 shenreng9qgrg132 http://