L2 Throughput

Download as pdf or txt
Download as pdf or txt
You are on page 1of 8

WHITE PAPER

ETHERNET DATA THROUGHPUT


EXECUTIVE SUMMARY
This paper provides an insight on the Ethernet frame, the data suppression and compression techniques available to
improve Ethernet frame throughput, and the differences between L1 and L2 throughput figures.
Increasingly, data throughput on Ethernet wireless links is being quoted at Layer 1 (L1) rather than Layer 2 (L2).
L2 throughput has traditionally been used to indicate Ethernet throughput performance. Throughput, along with latency,
frame loss and back-to-back frames, is provided using an automated RFC 2544 test on dedicated Ethernet/IP test
equipment. RFC 2544 testing provides an independent industry-standard means to fully validate and benchmark an
Ethernet network connection.
Other means to measure throughput include PC-based TCP or UDP test suites. These operate at Layer 4 (L4) and are
represented by products such as Iperf and Netperf. Their results can be highly variable and are not recommended for
validation testing. But in the hands of operators who are aware of their operation and limitations, they can provide a
quick and useful indication of L2 performance.
More recently, L1 throughput figures have appeared as a measure of raw bit-rate throughput. Unlike L2 throughput,
which provides figures based on the Ethernet frame size, L1 measures on the frame space; the Ethernet frame size plus
the inter-frame IFG and Preamble bytes. L1 figures are usually indicated for smallest, 64-byte frames to provide a
maximized L1 throughput figure. Some suppliers are using these maximized L1 figures for their products without
reference to the methodology, leading to exaggerated and misunderstood claims within an industry more used to
throughput figures provided under RFC 2544.
Throughputs can be improved using bit-saving data suppression and compression techniques. Non-essential framing
data on a link between L2 switches is removed at the input to the link, and then restored on exit, to provide an effective
means to increase throughput and hence data efficiency.

March 31, 2010

WHITE PAPER
ETHERNET DATA THROUGHPUT

TABLE OF CONTENTS

EXECUTIVE SUMMARY ........................................................................................................................... 1


Ethernet Framing ....................................................................................................................................... 3
L1 vs. L2 Throughput ................................................................................................................................. 6
L1 or L2 Throughput .................................................................................................................................. 7
Measurement Considerations.................................................................................................................... 8
RFC 2544..................................................................................................................................................................... 8
Other Test Methods ..................................................................................................................................................... 8

CONCLUSION ........................................................................................................................................... 8
Glossary ..................................................................................................................................................... 8

2 AVIAT NETWORKS

March 31, 2010

WHITE PAPER
ETHERNET DATA THROUGHPUT

ETHERNET FRAMING
IP data packets are encapsulated within Ethernet frames for transport over Ethernet networks. Figure 1
illustrates Ethernet framing of an IP packet for standard and VLAN tagged (802.1Q and 802.1Q-inQ)
frames.

Figure 1. Ethernet frame

Description

Bytes

IFG

Inter-Frame Gap

PRE

Preamble: Synchronization + SFD


Consists of seven preamble synchronization bytes (alternating ones and zeros),
plus one start of frame delimiter (SFD) byte.
Destination Address 6 bytes
Source Address 6 bytes

MAC DA and
MAC SA
(Media Access
Control)
L/T

12
8
12

Length/Type

Q Tag

VLAN 802.1Q Tag

Q-in-Q Tag

VLAN 802.1Q-in-Q Tag

Packet

Includes:
IP Header 20 bytes typical
TCP header 20 bytes
Application data max 1460 bytes (std packets)
Frame Check Sequence

FCS

46 to 1500

4
Total Frame Space: Std Frame

84 to 1538

Total Frame Space: Q Frame

88 to 1542

Total Frame Space: Q-in-Q Frame

92 to 1546

Table 1: Ethernet Frame Content (802.3)

For each packet


sent, the IFG and

Standard IP data packets range in size from 46 to 1500 bytes. Each packet is encapsulated within
Ethernet framing, which adds frame address and check bytes, plus an inter-frame gap (IFG) and a
Preamble (PRE).

Preamble consume
20 bytes.

3 AVIAT NETWORKS

March 31, 2010

WHITE PAPER
ETHERNET DATA THROUGHPUT

The Ethernet Frame comprises the IP Packet plus MAC header (MAC DA and MAC SA),
Length/Type and FCS bytes. The number of error-free frames that can be sent over an Ethernet link
represents the L2 throughput for that link
The Ethernet Frame Space comprises the Ethernet Frame plus IFG and Preamble bytes. The
number of error-free frame-spaces that can be sent over an Ethernet link represents the L1
throughput for that link
The first items of interest are the IFG and Preamble. Together they add 20 bytes to the Ethernet frame
size and as overheads deliver no useful content data.
The Ethernet specification (IEEE 802.3) requires the IFG to be not less than 96 bits = 12 bytes
The 8 byte Preamble (Preamble + SFD) at the start of each frame heralds the arrival of the frame
This means that when an Ethernet frame is sent, 20 bytes are added for the transmission of the
frame
The next items of interest are the MAC destination address (MAC DA) and MAC sender address (MAC
SA). These sit within the Ethernet frame, and at 6 bytes each represent a 12-byte total. Again, they are
an overhead they deliver no useful content data.
This is where modern Ethernet link equipment, such as the Aviat Eclipse Packet Node, act to provide
throughput efficiencies. They do this through suppression of the IFG and Preamble, and compression
of the MAC DA and MAC SA. This is illustrated in Figure 2 for a standard Ethernet frame.

IFG + Preamble suppression and MAC Header compression particularly apply to rate
or bandwidth limited links within an Ethernet network, such as microwave links or
leased-line connections, where making most efficient use of available capacity is
paramount. They are typically applied over links between L2 switches where the
dedicated nature of the connection means theres no need to retain IFG and Preamble
bytes and the MAC address data. Instead, these overhead bytes are suppressed or
compressed at the sending end and then reinserted at the receiving end.

The standard
frames accepted at
the input are
reduced in size for
transmission over
the link and then
reinstated at the far
end.

Figure 2. IFG and Preamble suppression and MAC DA/SA compression

4 AVIAT NETWORKS

March 31, 2010

WHITE PAPER
ETHERNET DATA THROUGHPUT

Item

Description

Standard Ethernet frame. Frame size 64 to 1518 bytes. Frame space 84 to 1538
bytes.

Application of IFG and Preamble suppression. The 20 bytes are replaced by 4


bytes; 2 bytes each for transport channel start-of-frame and end-of-frame
delineation, and frame scrambling.
This 16-byte reduction in frame space represents a 23.5 percent throughput
improvement for 64-byte frames. For average-size 260-byte frames, it represents a
6 percent traffic throughput improvement. For 1518-byte frames, it represents a
1 percent improvement.

Addition of MAC header compression. The 12 bytes are replaced by 2 bytes to


support an address look-up-table across the link.
When this 10-byte reduction is added to the 16 bytes of IFG and Preamble
suppression, the total represents a 44.8 percent throughput improvement for 64byte frames. For average-size 260-byte frames, it represents a 10 percent traffic
throughput improvement. For 1518-byte frames, it represents a 1.7 percent
improvement.

Table 2. Frame suppression and compression action

Ethernet overhead
varies and is a
function of frame
size/distribution.

Lets look at the math to support this. We have assumed a 100 Mbit/s link capacity for simplicity, and
that the full data bandwidth of the link is available for payload transport. In practice, on Ethernet
wireless links where there is a need to match Ethernet framing to the air-link framing, a small
percentage of the payload will be consumed by the shaping and handshaking overheads required.
The number of frames sent per second can be calculated by dividing the link capacity in bit/s by the
frame space in bit/s. For the smallest 64-byte frame, the number of standard Ethernet frames sent per
second on a 100 Mbit/s link is: 100,000,000 / (64 + 20) x 8 = 148,809.5 FPS. Note again that the
frame space is the Ethernet frame size plus the 20 bytes of IFG + Preamble.
This is where suppression of the IFG and preamble bytes provides real value. Suppression reduces
the frame space by 16 bytes; from 20 to just 4 bytes. So for the same 64-byte frame, we can now send
through 100,000,000 / (64 + 4) x 8 = 183,823.5 FPS. The extra FPS sent compared to standard frame
transmission is 35,014, which represents a 23.5 percent increase.
With MAC header compression, the frame space reduces by a further 10 bytes. For the same 64-byte
frame, we can now send through 100,000,000 / (64 + 4 - 10) x 8 = 215,517.2 FPS. The extra FPS sent
compared to standard frame transmission is 66,707.7, which represents a 44.8 percent increase.

The average frame


size is about 260
bytes.

Figure 3 summarizes the percent throughput improvements for several frames sizes, including an
average-size 260-byte frame.
The smaller the frame size, the more frames it takes to fill the pipe. And with more frames, there are
more overheads, so overheads consume a proportionally larger slice of the available Ethernet
bandwidth. Hence, the smaller the frame size the greater the gains to be made by employing frame
suppression and compression techniques. Put another way, by reducing overheads more useful
(packet) data is sent inside each frame.
Frame
Size

Standard Frame
Frame
Space

FPS

Frame with IFG & Preamble


Suppression
Frame
FPS
FPS %
Space
Increase

Frame with IFG & Pre Supp. +


MAC Header Compression
Frame
FPS
FPS %
Space
Increase

64
84
148809.5
68
183823.5
23.5
58
128
148
84459.5
132
94697.0
12.1
122
260
280
44642.9
264
47348.5
6.1
254
512
522
23496.2
516
24224.8
3.1
506
1518
1538
8127.4
1522
8212.9
1.1
1512
Figure 3. Summary of throughput improvements by frame size for a 100 Mbit/s link

5 AVIAT NETWORKS

215517.2
102459.0
49212.6
24703.6
8267.2

44.8
21.3
10.2
5.1
1.7

March 31, 2010

WHITE PAPER
ETHERNET DATA THROUGHPUT

L1 VS. L2 THROUGHPUT
L1 throughput represents the total bit rate through an Ethernet user port based on frame space; the
Ethernet frame plus the full 20-byte IFG + Preamble.
Where a link does not support IFG + Preamble suppression and MAC Header compression, the L1
throughput represents the link throughput
Where a link incorporates these suppression and compression capabilities it can accept more
incoming frame spaces more bit/s than would otherwise be the case. This is frame-size
dependent smaller frames have a greater proportion of their frame space taken up by the IFG +
Preamble overhead so they have more to gain with the application of IFG + Preamble suppression
L1 throughput is sometimes referred to as the port utilization rate
L2 throughput represents the bit rate count on Ethernet frames through a user port. The count does not
include the IFG + Preamble bytes.
Throughput is highly frame-size-dependent on links that do not support IFG and Preamble
suppression
On links that do, the reduction in overhead from 20 to just 4 bytes improves L2 throughput on all
frame sizes. But smaller frames receive most benefit, resulting in L2 throughput being much less
affected by frame size
MAC header compression reduces the frame size; standard frames accepted at the input are reduced
in size for transmission over the link, and then reinstated at the far end. The 10-byte reduction benefits
both L1 and L2 throughput figures. Smaller frame sizes benefit most.
Figure 4 shows a comparison of L1 and L2 throughputs on a 100 Mbit/s link. From this, it can be seen
that:
Without IFG + preamble suppression or MAC header compression:
L1 port throughput matches the link capacity of 100 Mbit/s and is independent of frame size
L2 throughput is significantly affected by frame size. Larger frames are transported more efficiently
the smaller the frame the higher the proportion of IFG + Preamble overhead, and the lower the
throughput. Throughput on smallest 64-byte frames is nominally 76 Mbit/s (76 percent of link
capacity). On average size 260-byte frames, throughput is 93 Mbit/s (93 percent). On 1518-byte
frames, throughput is 98.7 Mbit/s (98.7 percent)
With IFG and preamble suppression:
The reduced overhead on frames sent over the link allows more frames to enter the link.
L1 port throughput increases to 123 Mbit/s for 64-byte frames, to 106 Mbit/s for average-size 260byte frames and to 101 Mbit/s for 1518-byte frames
L2 throughput increases to 94 Mbit/s for 64-byte frames, to 98.5 Mbit/s for 260-byte frames and to
99.7 Mbit/s for 1518-byte frames. Throughput is much less affected by frame size
The percent improvement in throughput for L1 and L2 compared with a standard frame (no
suppression) is 23.5 percent for 64-byte frames, 6.1 percent for 260-byte frames and 1.1 percent for
1518-byte frames

6 AVIAT NETWORKS

March 31, 2010

WHITE PAPER
ETHERNET DATA THROUGHPUT

With IFG and preamble suppression, plus MAC header compression:


L1 port throughput increases to 144.8 Mbit/s for 64-byte frames, to 110 Mbit/s for 260-byte frames
and to 101.7 Mbit/s for 1518-byte frames
L2 throughput increases to 110 Mbit/s for 64-byte frames, to 102 Mbit/s for 260-byte frames and to
100.4 Mbit/s for 1518-byte frames
The percent improvement in throughput for L1 and L2 compared with a standard frame (no
suppression or compression) is 44.8 percent for 64-byte frames, 10.2 percent for 260-bytes frames
and 1.7 percent for 1518-byte frames
160,000,000
Standard Frame L1 Port Speed

150,000,000
140,000,000

IFG + Pre Suppression L1 Port


Speed

130,000,000

Bits/s

120,000,000
110,000,000

IFG + Pre Suppression plus


MAC Compression L1 Port
Speed

100,000,000

Standard Frame L2 Throughput

90,000,000
IFG + Pre Supression L2
Throughput

80,000,000
70,000,000
60,000,000
0.0

500.0

1000.0

1500.0

IFG + Pre Suppression plus


MAC Compression L2
Throughput

Frame Byte Size

Figure 4. L1 and L2 Mbit/s as a function of frame size

L1 and L2
throughput figures
are now in use.
Check to see which
is being

L1 OR L2 THROUGHPUT
Figures quoted at L1 are universally based on 64-byte frames to provide a maximized indication of
bits-per-second throughput. L1 figures are increasingly being used within the industry.
Layer 2 throughput figures have traditionally and more realistically represented Ethernet throughput, as
they look only at the Ethernet frame; IFG and preamble bytes are ignored. Measurement practice is
specified in RFC 2544 to provide a standardized test methodology for the industry.

represented.

7 AVIAT NETWORKS

March 31, 2010

WHITE PAPER
ETHERNET DATA THROUGHPUT

MEASUREMENT CONSIDERATIONS
Data throughput refers to the maximum amount of data that can be transported from source to
destination with zero errors or lost frames. As a test, it measures the rate-limiting capability of an endend Ethernet connection.

RFC 2544
Throughput is normally benchmarked using an automated RFC 2544 test:
RFC 2544 defines a specific set of tests to measure and report the performance characteristics of
network devices. It provides a standardized test methodology for the industry
It tests for throughput, latency, frame loss and back-to-back frames
The test suite supports seven predefined frame sizes (64, 128, 256, 512, 1024, 1280 and 1518
bytes) to simulate various traffic conditions
RFC 2544 testing is supported on all professional Ethernet test equipment
L1 throughput can be calculated using the L2 frame rate data provided from an RFC 2544 test. L1
throughput = frame rate FPS x frame space bytes x 8, where the frame space is the frame size plus
the full 20 bytes of IFG + Preamble.

OTHER TEST METHODS


Application test procedures using software such as Iperf or Netperf are frequently used to measure
throughput. These are layer 4 networking evaluation tools and test figures can be misleading unless
users are familiar with their operation, limitations and the results that can be expected. Results must be
treated with caution raw results can be pessimistic by 10 percent or more.
For information on Iperf testing, refer to the Aviat application note: Iperf Testing of DAC ES
Performance.

CONCLUSION
Both L1 and L2 throughput figures are being used within the industry. If the measurement methodology
is not stated, ask.
RFC 2544 is the preferred validation tool for Ethernet network testing.
IFG + Preamble suppression and MAC Header compression significantly improve data transport
efficiencies.

GLOSSARY
IETF
MAC
RFC
RFC 2544
VLAN

Internet Engineering Task Force. IETF recommendations, Internet standards or network protocols
are published as RFCs.
Media access protocol.
Request for comments.
An IETF recommendation that specifies test criteria to characterize Ethernet link performance.
Virtual LAN. IEEE 802.1Q tagging mechanism.

WWW.AVIATNETWORKS.COM
Aviat, Aviat Networks and Aviat logo are trademarks or registered trademarks of Aviat Networks, Inc.
Aviat Networks, Inc. (2009 2010). All Rights Reserved.
Data subject to change without notice.
_w_Ecli_Ethernet_Thruput_UNIV_31Mar10.doc
8 AVIAT NETWORKS

March 31, 2010

You might also like