30812016 Network model for ITU

advertisement
Telecommunications Industry Association
Lake Buena Vista, FL
December 8 - 9, 2008
TR-30.3/08-12-016
Document Cover Sheet
Project Number
PN-3-3175-RV2
Document Title
Network model for G.8261
Source
Anue Systems
Contact
Name: Chip Webb
Complete Address: 9111 Jollyville Rd
Austin, TX 78759
Distribution
TR-30.3
Intended Purpose
of Document
(Select one)
x
Phone: 512-527-0453x102
Fax:
Email: cwebb@anuesystems.com
For Incorporation Into TIA Publication
For Information
Other (describe) -
The document to which this cover statement is attached is submitted to a Formulating Group or
sub-element thereof of the Telecommunications Industry Association (TIA) in accordance with the
provisions of Sections 6.4.1–6.4.6 inclusive of the TIA Engineering Manual dated March 2005, all of
which provisions are hereby incorporated by reference.
Abstract
Like TIA-921, ITU-T G.8261 seeks to model delay and loss in packet networks. This document describes
a model used for G.8261 and shows comparisons between that model results and actual laboratory
measurements.
4/8/2015
v1.0 - 20050426
Anue Systems, Inc.
www.anuesystems.com
1
Problem statement

G.8261 currently describes a way (in Appendix VI) to test
sync Interworking Functions by building a testbed.
But -- it’s not repeatable.
4/8/2015
Anue Systems, Inc.

www.anuesystems.com
2
Initial results match
Input profile
4/8/2015
Measured result
Anue Systems, Inc.
www.anuesystems.com
3
A test bed

4/8/2015
Measure delay under G.8261 Appendix VI test conditions…..
Anue Systems, Inc.
www.anuesystems.com
4
Model-based impairment profiles


Build a bottom-up model of each network element
Construct test scenarios by connecting together
various model elements and run a simulation


Still has shortcomings



4/8/2015
Validate end-to-end results
As network changes, model params must also change
New technologies may require new model elements
Harder to make a general model than to measure one
sample
Anue Systems, Inc.
www.anuesystems.com
5
Why create a software model?

Creating a model and validating it against
real networks provides valuable insight to
guide further work




4/8/2015
Which device parameters are most important
What metrics work best
What’s the best way to test (interop/conformance)
Understand tradeoffs in system deployment
Anue Systems, Inc.
www.anuesystems.com
6
Discrete event simulation of test bed..




One way to model is to use a discrete event
simulator
Develop models for the switches and
dummy traffic generators.
Anue is developing one such model for
MEF18 testing.
This is just the beginning of modeling

4/8/2015
Further refinements are possible
Anue Systems, Inc.
www.anuesystems.com
7
Switch model




4/8/2015
Three main blocks in the switch model
 Ingress, Queuing, Egress
Ingress block:
 Each input port has one. Makes forwarding decisions (L2/L3)
 There’s no contention here.
 Introduces delay (store/forward or cut-through)
Queuing block:
 One or more queues per output port. Holds packets till sent out.
 Contention can happen here. Queue has limited size (shared)
 Queue memory allocated in N-byte chunks.
 May implement congestion avoidance (e.g. WRED)
 Introduces queuing delay and packet loss.
Egress:
 Each output port has one. Services queues at the port’s bit rate.
 If multiple output queues, contention can also happen here (e.g. Strict
Priority, WRR, WFQ)
Anue Systems, Inc.
www.anuesystems.com
8
Model assumptions

Initial focus is queuing delays






4/8/2015
In the forward direction only
All switches are non-PTP capable (asynch)
No priority or congestion avoidance
Wire delay is constant
Assume each switch has at least one flipflop domain transfer
Ignore oscillator noise
Anue Systems, Inc.
www.anuesystems.com
9
Model parameters



Ten switches (based on G.8261)
Dummy load is Traffic Model 2
Queue size is 64k bytes

Allocated and deallocated in 64 byte chunks

All links are gigabit
Measure delay 1000 times per second

Model outputs



4/8/2015
Packet delay
Packet drop
Anue Systems, Inc.
www.anuesystems.com
10
But can’t realistically simulate all packets

A 24-hour simulation of a 10 hop network
built out of GE switches and operating at
50% load with 1400 byte (avg) packets
represents about 40 Billion packets.

That’s half a petabit.
If you watched HDTV for two years straight,
without sleeping, it would use about that
many bits.

4/8/2015
Anue Systems, Inc.
www.anuesystems.com
11
Example model results (20% load TM2)
S/W Model Results
4/8/2015
Reference (test bed)
Anue Systems, Inc.
www.anuesystems.com
12
Simulation Results MEF18, 6.2a
Based on G.8261 Appendix VI Test Case 3
4/8/2015
Anue Systems, Inc.
www.anuesystems.com
13
Simulation Results MEF18, 6.6a
Based on G.8261 Appendix VI Test Case 5 (congest 100s)
4/8/2015
Anue Systems, Inc.
www.anuesystems.com
14
Download