Docstoc

Measuring the use of the hybrid network

Document Sample
Measuring the use of the hybrid network Powered By Docstoc
					              GLIF, the Global Lambda Integrated Facility
          Measuring the use of the hybrid network




Erik-Jan Bos                     Minneapolis, MN, USA
Managing Director SURFnet        February 14 & 15, 2007
                                                        2
The hybrid network enables a range of services

• Previous NREN generations offer only IP services
• The hybrid network offers services at different
  levels:
   – IP
   – Sub-lambda lightpaths (GFP-F / VCAT / LCAS)
   – Full lambda lightpaths
• Using modern network equipment and control
  plane, the NREN can make lightpaths more dynamic
   – Easy configuration for static lightpath services
   – User controlled lightpaths: scheduled or dynamic
                                                          3
New metrics needed to measure usage in hybrid network

• Previous generations reported primarily on IP traffic
   – Metrics, measurement methods
     and tools are well known
   – Byte count provides sufficient
     information for many purposes
   – Graphs are easy to understand
• Lightpaths are different
   – Connection oriented
   – Static, scheduled, or dynamic
• How to report on services, and on the network as a
  whole?
                                                                            4
Start by mapping services to network capabilities

           Network capability            Service

Layer 3     IP Routing                       IP


Layer 2             Ethernet
                   aggregation       Ethernet VLAN


Layer 1   Sonet framed (GFP/VCAT)   Lightpath (sub-λ)



                 Lambda             Lightpath (λ rate)
Layer 0
                    Fiber                Dark fiber


                                                         Service not currently
                                    SURFnet service
                                                         offered by SURFnet
                                                    5
Who is it for?

• General public and policy makers:
   – need to justify funding


• Connected organizations:
   – need to understand use of the service
   – need to justify costs


• Network operator:
   – need to allocate costs (at least internally)
   – need to plan network build-out
   – need to make routing decisions
                                                           6
Different stakeholders, different levels of complexity


                               General public and
                               policy makers




                                       Connected
                                       organizations




                                                Network
                                                operator
                                                                      7
General public & policy makers: justify spending money

• This group requires simple metrics: easy to explain,
  easy to understand


• Clear link to the value delivered through the network




                                      Source: SURFnet annual report
                                                                             8
General public and policy makers: keep it simple

• Simply count all the bytes transported by the
  network
     30000
                                                            Lightpaths:
     25000                                                     bytes
                                                            transported
     20000

     15000                                                  IP bytes
                                                            outgoing
     10000


       5000

            0                                               IP bytes
                                                           incoming
                     05




                                               06
                                               06
                                                5
                                               05




                                               06
                              5




                                               06
         05




                                              06
                                             00
                            00
                   20




                                            20
                                            20
                                            20




                                            20
       20




                                            20




                                            20
                                          r2
                          t2




                                         ry
                  ne




                                        ne
                                        er




                                        er
                                        st
      ril




                                        ril
                                       be
                           s




                                      ua
    Ap




                                     Ap
                         gu




                                      gu
                                     ob




                                     ob
                Ju




                                     Ju
                                    m


                                   br
                       Au




                                  Au
                                   ct




                                   ct
                                  ce


                                Fe
                                 O




                                 O                  (hypothetical example)
                               De
                                                                                                                                                                                                                 9
General public and policy makers: keep it simple

• …and report total capacity provided for all services

                         500
                                                                                                                                                                                                 Dynamic
                         450                                                                                                                                                                    lightpaths
                                                                                                                                                                                            (bandwidth x % of
                         400                                                                                                                                                                   time active)
                         350
      Gigabit / second




                         300

                         250
                                                                                                                                                                                             Total bandwidth
                                                                                                                                                                                             static lightpaths
                         200

                         150
                                                                                                                                                                                                Transport
                         100                                                                                                                                                                  capacity for IP
                         50

                           0
                                                                                                                                                                                        (hypothetical example)
                                                                                                                         jun-06
                                                                                                                                  jul-06
                                                          nov-05




                                                                                                                                                                      nov-06
                                                                            jan-06
                               aug-05




                                                                                                                                           aug-06
                                                                   dec-05




                                                                                              mrt-06
                                                                                                       apr-06
                                                                                                                mei-06




                                                                                                                                                                               dec-06
                                        sep-05
                                                 okt-05




                                                                                     feb-06




                                                                                                                                                    sep-06
                                                                                                                                                             okt-06
                                                             10
Connected organizations: understanding service delivered

• Want to understand price/performance
• Want to pinpoint capacity bottlenecks
• May want to charge internal user groups
                                         My organization:
                                       Interface A by week
• Therefore:
   – More detail, less aggregation
   – Easy access to information
   – Different timescales
                  My organization:
               Interface A by minute
                                                                        11
Connected organizations: more detail, less aggregation

• Separate reporting per service delivered:
   – IP:
      • Packets and octets, peak and average, at different timescales
      • Packet loss
      • Flow monitoring above the IP layer?
   – Static lightpath:
      • Interface capacity
      • Transport bandwidth delivered to the interface
      • Octets transported?
   – Dynamic lightpath:
      • Interface capacity
      • Transport bandwidth requested and granted
      • Percentage of time active, number of requests, number denied
      • Octets transported?
                                                       12
Network operator: accounting and charging

• Most NRENs have very simple pricing schemes
   – no need for detailed charging per customer


• However, they still need to make pricing decisions
  based on actual facts


• Therefore they need to understand:
   – the use of each service
   – the resulting claim on resources
   – and the cost of each resource
                                                                        13
Network operator: mapping to resource usage

• Report on the services delivered
   – Same metrics as provided to connected organizations


• Infrastructure
   – Static resources used by each service
      • e.g. IP connection using n VC-4’s on a given link
   – Variable resources used by each service
      • e.g. dynamic lightpath using n VC-4’s on average, m peak, x %
        of the time
                                                                                               14
Network operator: mapping service to use of infrastructure
                                                                      Sim
                                                       IP service
                                                                            pli
                                                     uses 0.2% of              fi   ed
IP service                                              available
 uses one                                            uplink at peak
                                                                                       e   xa
 fiber pair             IP service                                                           mp
                        uses 7 VC-                                                             le
                        4’s on this
                          lambda

        IP
        Lightpath                                                           Lightpath



 Scheduled
  lightpath
service uses
one fiber pair

                    IP service uses
                    7 VC-4’s on one
                                      Scheduled lightpath
                     of 8 lambda’s
                                       service uses 7 VC-
                                      4’s on same lambda,
                                          1% of the time
                                                                    15
Network operator: network planning and design

• NREN needs to pinpoint bottlenecks…
   – preferably before they become bottlenecks!
• … at each layer in the network
   – but not in isolation: each layer is related to the others
• … at each device and each link
   – but not only at the local level: bottlenecks may be resolved
     by re-routing
• … at different timescales
   – busiest hour, busiest day...
                                                                                                            16
  Network operator: local capacity monitoring at each layer
                                                                                   Sim
                                                                10 Gbps router port:     pli
                                                                                            fi   ed
                                                                 average 1 Gbps,
                                                                   peak 8 Gbps                      e   xa
                             Lambda:                                                                      mp
                       7 VC-4’s used for IP,                                                                le
                    14 reserved for LP service,
                           43 available

           IP
           Lightpath                                                                     Lightpath




       Cable:
two fiber pairs used,
   two available
                                                Fiber pair:
                                            3 lambda’s used,
                                               1 available,
                                            space for 68 more
From local to global:                                                                                   17
build up an end-to-end traffic matrix at each layer
            end-to-end


        IP Traffic (Mbps/sec)
        (peak day, peak hour)
                   University A       University B     Upstream         ...
  Organization     Site:              Site:
         Site      A1      A2         B1
                                         Lightpaths in use (VC-4’s): static, dynamic
  University A
                                         (peak day, peak hour)
         Site A1   -       0,2       0,2               0,2
         Site A2   0,2     -         0,2             University A
                                                       0,2          University B   IP Routing   GEANT
  University B                   Organization        Site:          Site:
                                        Site         A1       A2    B1
         Site B1   0,1     -         0,1                0,5
                                 University A
                                         Site A1     -        -     2 / 14         14           14
  Upstream         0,3     -          0,6Site A2     - -      -     -              2
                                 University B
  ...                                    Site B1     2/7      -     -              7            7


                                 IP Routing          14 / 7   2     7              -            -



                                 GEANT               14             7
                                                        18
Network operator: capacity planning for lightpath services

• IP is packet switched, but lightpaths are circuit
  switched - just like telephony

• The telephony world has a good capacity model: the
  Erlang model

• So can we use the Erlang model to predict capacity
  requirements in the lightpath environment?
                                                                  19
The Erlang model will not work for lightpaths:
at least not in the near future


• The Erlang model assumes:
    – Stochastic distribution of arrival times
    – Stochastic distribution of duration
• Why does this work for telephony:
    – Very large number of potential callers, each contributing
      very little, with independent reasons for calling
       • All calling at random times   Poisson distribution
       • Negative exponential distribution of duration
• Why does this not work for lightpaths
    – Small number of potential users at this time
    – Limited number of applications at this time
                                                                        20
Accurate forecasting of lightpath use will not be possible

• New service        no data to extrapolate from
• Small number of applications     large impact from
  use patterns of each application
• Small number of users             large impact from a single
  user decision
• Example: imagine an organization offering a remote
  backup service across scheduled lightpaths
      • What time does a scheduled backup start?
        Probably at 20:00 or 0:00
      • How long is the backup scheduled for?
        One hour, or two.
      • Fairly predictable pattern, but a change in backup strategies
        can disrupt the pattern
                                                                        21
Therefore: augment data from the network with knowledge f



• Measure the use of resources in the network
                        This lambda:
                    7 VC-4’s used for IP,
                 14 reserved for LP service,
                        43 available                    Actual use
                                                        of dynamic
                                                         lightpaths

                                                   Reserved
                                                 for dynamic
                                                   lightpaths
                                               Static lightpaths
• Show historic data to
  see first order trends                       Used for IP

                                                    Actual IP traffic
Questions?
                                                         23
Potential metrics, layer by layer

• Layer 0: fibers and lambda's
   – kilometers of cables and fibers
   – fibers lit
   – lambda’s available in WDM systems
   – lambda’s in use and total bandwidth, by use type:
      • used for higher network layers
      • used as static lightpaths
      • reserved for dynamic lightpaths
   – use of dynamic lambda lightpaths:
      • bandwidth x hours of use
      • number of set-up requests
      • experienced blocking probability
   – octets transported?
                                                            24
Potential metrics, layer by layer

• Layer 1: optical switching
   – number and size of lightpaths provided, by use type:
      • used for higher network layers
      • used as static lightpaths
      • reserved for dynamic lightpaths
   – use of dynamic sub-lambda lightpaths:
      • bandwidth x hours of use
      • number of set-up requests
      • experienced blocking probability
   – octets transported?
                                                      25
Potential metrics, layer by layer

• Layer 2: Ethernet aggregation
   – number of interfaces
   – lightpath bandwidth
   – aggregation ratio
   – number of frames transported, dropped, errored
   – number of frames, by upper layer protocol
   – octets transported
                                                              26
Potential metrics, layer by layer

• Layer 3: IP
   – number of interfaces
   – transport capacity per interface (lightpath bandwidth)
   – packet loss
   – packets transported: IPv4/IPv6, unicast/multicast
   – octets transported: IPv4/IPv6, unicast/multicast
   – protocols above IP (netflow)

				
DOCUMENT INFO
Shared By:
Tags: Hybrid, network
Stats:
views:18
posted:8/8/2011
language:English
pages:26
Description: Hybrid network is made up of different topologies (such as ring and star) to build the network.