Introduction

This study presents examples of distributions of Ethernet packet sizes, called EMIX profiles, in the Metro Ethernet network. The EMIX concept has been defined in a separate technical note published by the Ethernet Academy Web Site. Ethernet service flows are indisputably composed of packets of different sizes (reflecting, among other factors, the mixture of protocols used in transported services). However, more often than not, the compositions of these flows are unknown.

Network services are usually verified with test flows that consist of packet sizes of an arbitrary composition (following the RFC 2544 recommendation ) (Bradner 1999, Morton 2011). However, tests conducted with these types of flows will not accurately reflect the performance of the service in the production network. This is particularly true when the service is provided over a shared network with a complex traffic congestion-management design. The tests that employ flows with arbitrary patterns of packet sizes are more suited for characterization of networking hardware in the lab environment than they are for service performance over the production network . The shortcomings of these tests were recognized by the IP community, which formulated the concept of IMIX to address this issue. In response to similar challenges, the Ethernet community developed the idea of EMIX, which provides packet size profiles for Ethernet service flows.

Data

In this technical note, the size of frames in the EMIX is defined by the service frame size, as specified by MEF[7] . The MEF concept of the service frame excludes the Preamble and Interface gaps, but includes the FCS field[8].

The data for the study reported in this note were collected over several thousand UNIs over more than 12 months in different Metro regions across the Continental US. No special equipment was used for the collection of data. As well, no information about the specific type of services was accessible. The only data available for the study were Layer 2 packet size cumulative statics collected at UNI interfaces, in six packet size ranges of 64 bytes, 65–127 bytes, 128–255 bytes, 256–511 bytes, 512–1023 bytes, and 1024–1518 bytes. The collected packet ranges correspond to the router statistics collection capabilities.

The packet statistics were collected in six sets – A, B, C, D, E, and F – which reflected six different, arbitrarily selected combinations of UNIs. Detailed information about each set cannot be provided due to the limitations of the collection method and confidentiality restrictions. Furthermore, no systematic study of traffic patterns over time or geographic location has been conducted. Thus, statistically valid analysis of data could not be performed . The presented EMIX statistics should be perceived more as a snapshot of traffic flows for the selected time and location, rather than as accurately defined, statistically significant, universally defined flow patterns.

Results

Table 1 and Figure 1 present the EMIX for six differentiated sets and Figure 2 provides all EMIX statistics in a single graph.

Table 1. Combined EMIX profiles for all sets

(1)- Set A ; (2) Set B;(3) Set C; (4); Set D; (5) Set E; (6)- Set F.
Figure 1. EMIX sets.

Figure 2. Combined EMIX profiles for all identified UNI sets [10]

Table 2 presents the differences between specific EMIXes with respect to the percentage of packet size in each packet size class. For all EMIXes, the differences range from 8% to almost 17%. However, when two sets – A and F – are eliminated, one may venture the observation that most of the sampled EMIXes differ by less than 5% from each other (the exception is the 65–127 interval). These differences are also presented in Figure 3.

Due to the limitations of the data collection process, statistical analysis of the collected data could not be conducted.

Table2. Differences between EMIX compositions [11]

Figure 3. Differences in EMIX

Observations and Conclusions

The results of this study led to several interesting observations about EMIX composition in the Metro Ethernet network:

The EMIX profiles in the reported cases do not have a uniform packet composition.

• The EMIX profiles, with the exception of the subset profile A and F, are similar within a 5% difference.
• Most of the EMIX profiles are composed of packets predominantly below 511 Bytes; these packets constituted 64% or more of every EMIX observed in this study.
• The packets between 1024 and 1518 bytes, in most of the observed cases, constituted 25% or fewer of the packets in the flows.
• The packets between 256-1023 bytes constituted slightly more than 10% of all packets in the flows.

In the light of presented data, the recommended tests of the actual services should involve flows with varied packet sizes characteristic of the specific service. In the absence of the specific information about the service, the EMIX examples provided in this paper could be used for these tests. The authors hope that the observations discussed in this technical note will be included in the future updates to the ITU Y.1564 document (Y.1564 2011).

Reference

MEF 10.2. Ethernet Services Attributes Phase 2. October 2009.
ITU-T, Y.1564. Ethernet service activation test methodology. March 2011.
Bradner, S. and J. McQuaid, "Benchmarking Methodology for Network Interconnect Devices", RFC 2544, March 1999.
Morton, A.. IMIX Genome: Specification of variable packet sizes for additional testing. Draft-morton-bmwg-imix-genome-01. Expires: August 8, 2011.

© Verizon 2011, All Rights Reserved. © Ethernet Academy 2011, All Rights Reserved
All information provided without warranty and subject to change without notice.

Written by:

Roman Krzanowski

 

Frank Shannon

 

TrackBack URI for this entry

Comments (0)

Subscribe to this comment's feed