Genesis: TCP and BGP Simulation under SSFNet

1
Progress in the Overall Project PIs C.
Carothers, S. Kalyanaraman, B. Sikdar, B.
Szymanski

• On-line network modeling and simulation Genesis
  extended to NS, SSFNet and GloMoSim with UDP and
  TCP traffic, BGP implementation is underway
• Network modeling and abstraction focus on
  wireless traffic characterization and generators
• Experiment design and integration traffic
  engineering and routing under OSPF and load
  balance optimization for BGP

Experiment Design
Network Abstraction And Decomposition
Parallel Discrete Event Simulation
Performance
Processor 1
Processor 2
Domain 2
Domain 1
Parameter 2
Parameter 1
P1min
P1max
Processor 3
Processor 4
Current operating point
Domain 3
Trial operating point triggering simulation
router
Link, simulated at packet level
Models of Inter-domain flows
Links crossing processor Boundary may cause
rollback
All three trial points can be Evaluated
concurrently
2
Genesis Progress in Distributed On-Line Network
Simulation

• PIs Bolek Szymanski and Chris Carothers
• RAs Anand Sastry, Yu Liu, Kiran Madnani
• Rensselaer Polytechnic Institute Troy, NY
• http//www.cs.rpi.edu/szymansk/sonms.html
• email szymansk, chrisc, sastra, liuy6, madnak
• _at_cs.rpi.edu

3
Genesis Distributed On-Line Network Simulation

• Space decomposition partition large network into
  disjoined individual domains, each simulated
  independently and concurrently with others.
• Time decomposition partition simulation time
  into separate intervals, each interval iterated
  over until all domain simulators converge.
• Synchronization exchange packet delay and loss
  information on flows originated externally to
  each domain at the end of each interval
  simulation (iteration). Message passing via
  sockets is used in farmer-worker parallel
  architecture.
• Basic domain simulation is currently implemented
  in
• NS, SSFNet and GloMoSim for UDP and TCP traffics.

4
Multi-phase Link Proxy Setup In SSFNet
5
Link Proxies for Transit TCP Traffics
6
Simulation Convergence Improvements
Implemented dynamic checkpoint intervals based on
the simulation conditions. Slow-start algorithm
is applied to the freeze intervals Start
with a small value of the initial freeze
interval, If no go-back occurred in
interval n, then freezen1tconst else
freezen1/2
Go back condition aggregated traffic comparison
is used to minimize the effect of small number of
traffic changes
7
Measurements on 64-node Network Configuration
Some Traffic (data flow) Comparisons
Results from 64-node network simulations
8
Farmer-Worker Architecture in GloMoSim

• Each domain has a maximum of eight neighbors - at
  each freeze, each domain will communicate to at
  most 8 neighbors. In our design, we have direct
  peer-to-peer communication between the domains
  (workers).
• In order for workers to synchronize between
  freezes we implement a server (farmer).
• The design is as follows
• Each domain repeats till the end of simulation
  the following statements
• 1. Send up to 8 messages to all its neighboring
  domains.
• 2. Receive logs from all of its neighboring
  domains.
• 3. Check for conflicting actions of each of its
  nodes in border regions and this node proxy in a
  neighboring domain using the received logs.
• 4. If a conflict exists, then send signal 1 to
  the farmer and go-back to re-simulate the current
  freeze interval, else send signal 0 to the
  farmer.
• 6. Receive sync_message from the farmer if
  sync_message 1 then go-ahead to the next
  interval,
  
  else goto step 1.
• The server repeats till the end of
  simulation the following statements
• 1. Receive the signals from all domains
  and send the sum of all signals to all domains as
  sync_message.

9
Measurements and Future Work
Future Work

• Interoperability between GloMoSim
• and different network simulators
• (NS, SSFNet)
• Proposed methodology Adding a
• new application type in GloMoSim
• which mimics the semantics of a
• flow from a wired network.
• Modeling the characteristics of the
• wireless network sources to create
• more realistic traffic generators.

10
Goals of BGP Simulation Under Genesis
Goals BGP has been increasingly used for some
forms of traffic engineering. Our goal is to
provide a novel outbound load-balancing
technique using BGP LOCAL_PREF settings and
aided by online simulation. Outbound means each
AS has several links and we want to distributing
traffic to theses links such that a complex set
of objectives are met. These objective functions
will represent more realistic needs of large
networks like those of ISPs and defence networks
than what is done today. BGP under Genesis To
support simulation in which BGP changes impact
background traffic while preserving speed and
scalability we use Genesis approach. Our goal was
to port BGP to Genesis using SSFNET, and to
measure the scalability and speed of BGP
simulation under Genesis.
AS 4
AS 1
AS 2
AS 5
AS 6
AS 3
BGP router
11
Farmer/Worker Support for BGP on Genesis
B - BGP Speaker
F - Farmer
Here B1, B2, B3 send Identification message to
the Farmer followed by their Update Times. Their
update times are broadcast to the other BGP
Speakers through the farmer. This process enables
the Send and Receive Synchronization between BGP
Speakers. All messages are sent between the
Famer and the BGP Speakers. No messages are
exchanged between the BGP Speakers themselves.
12
Traffic Measurements for a 4-domain Network
The graph shows a route update bought about by
bringing down Link 1. Traffic was then routed
through Link 2 as indicated by the increase in
the number of packets. This update was at
Simulation time of 40sec.
13
Future Work
Support for State Saving When per-flow statistics
are collected, route information and update times
have to be stored for rollback support in
Genesis. These stored data will ensure quick
convergence to the solution by the domains
enabling fast completion of each iteration and
yielding high simulation speed. Integration with
TCP Support for link proxies to enable use with
Genesis TCP Sources. Performance
Comparison Compare performance to existing
non-BGP simulations. Also compare the Genesis
system with integrated BGP to other systems with
BGP Speakers and TCP sources. This comparison
matrix will involve the following Scalability
Larger networks with 64-256 nodes and multiple
BGP Speakers in each domain. Efficiency Compare
simulation times to single domain systems with
the same number of nodes.
14
Impact of IEEE 802.11 MAC on Traffic
Characteristics

• Omesh Tickoo and Biplab Sikdar
• Department of ECSE
• Rensselaer Polytechnic Institute
• Troy, NY 12180

15
The 802.11 MAC Protocol

• CSMA/CA
• Exponential Backoff in the presence of collisions
• Channel reservation through RTS and CTS messages

16
Modeling Inter-Arrival Times

• Backoff counter
• Residual time for each host
• No. of slots till next transmission
• PMF of the inter-arrival times
• where

17
Packet Inter-arrival times (TCP packets)
18
Packet Inter-arrival Times (All-packets)
19
TCP Packet Inter-arrival Times
20
Second Order Scaling (router)
21
Second Order Scaling (source)
22
Conclusions

• Analytic model for the impact of 802.11 MAC on
  inter-arrival times
• Inter-arrival times show a multi-modal
  distribution (at short time scales)
• At higher time scales inter-arrival times have a
  heavy-tailed distribution
• Individual sources are mono-fractal while traffic
  at the routers is not

23
Minimizing Packet Losses in BGP Outbound Traffic

• Online Simulation is a general technique that can
  be applied to optimize a wide-range of complex
  objectives
• APPLICATIONS
• Minimization of packet losses by optimizing OSPF
  weights
• Load-balancing outbound traffic in BGP
  environment
• Minimizing packet loss in BGP environment
• Stabilizing average RED queue size under varying
  load, round-trip-times, etc.

24
Objective

• Minimize packet loss at the outgoing interfaces
  of AS under consideration
• Min S lÎLPdropl
• L is the set of outgoing links from the AS
• Pdropl is the fraction of packets lost at
  interface l
• Load-balance outbound traffic
• Min S lÎL(rl - rtarget)2
• rl is the utilization of the link l
• rtarget is the target utilization
• Define Load-Balancing Factor (LBF) S lÎL(rl -
  rtarget)2
• Subject to constraints such as
• Do not change paths too frequently
• Do not choose paths that exceed shortest path by
  a certain factor
• Prefer paths based on history of availability
• Optimise only the paths to hot destination
  prefixes
• Do not override ASs existing policies

25
Simulation Results
30 AS, 90 Routers, 90 Hosts