App Notes

June 16

Ethernet Testing

Ethernet can be monitored for certain errors while the network is in production, but to assure network

performance prior to initial turnup or to diagnose a problem, intrusive testing is required. A BERT, or Bit

Error Rate Test can be run to determine if a link is error free as well as to establish available bandwidth

or throughput. An Ethernet Test Set will run a packetized random pattern to simulate Ethernet traffic

over a connection. This is done from one set to another or from one set to a loopback on another port.

A full Duplex bidirectional test using two sets is the preferred method since when using a loopback, the

direction of error initiation cannot be determined.

Before running a BERT, it is best to run a PING test between the two test sets conducting the BERT. This

makes sure there is actually a connection between the ports under test and that the test sets are

properly setup.

A typical test set can run a BERT at the physical, MAC address and/or IP layer. For the MAC layer, each

test set must be assigned respective source & destination addresses; the same rule applies for the IP

layer as well. A test set will also allow the user to set frame size and the amount of bandwidth to be

utilized. Test results from a BERT include collisions, frame count, errors, type of errors and bandwidth

utilization.

An example of a typical test scenario is to set the packet size at 100 bytes with utilization at fifty percent

for a period of fifteen minutes. On a 100mbps link this will mean there will be significant overhead

because the packet size is relatively small, and the utilization will be 50mbps, which is a reasonably

stressful testing benchmark.

A suite of tests that includes a BERT is defined by RFC2544. This series of tests changes the packet size

and bandwidth in a ramped mode. Be aware that this test will fully load the test link so it is not intended

for use on a production network.

Below is a glossary of terms and typical benchmarks for testing.

IP Network Testing Guidelines

Utilization – should average < 40%; note long bursts of traffic >60%.

Collisions – should not average > 5%. ≤ 1% is typical.

Errors – there is no maximum or minimum specified for Ethernet in 802.3, but a small number of errors

can mean a problem is starting. A large number of errors mean a problem already exists.

Subnets – one per network.

Ping – use this at network setup to measure and baseline response time and hop counts.

Bert Tests – a good traffic sample to run a BERT with on a production network in order to determine

traffic related problems on a live network is a frame size of 100 bytes.

Packet Delay – data, < 4ms one way, < 8ms round trip for toll grade voice ; < 100ms for data.

Jitter - < 50ms data; < 2ms for toll grade voice.

Packet Loss - < 1%

Frame Loss Rate - < 6.25e 10/7th

Svc Availabilty – 99.9%

MTTR (Mean Time To Repair) - <4 hours

9:59 PM GMT | Read comments(0)

Network Products