Data throughput on an Ethernet network
What's the best way to squeeze that last ounce of performance from your good old Category-5 Ethernet network?
01 October 2004
Getting the most out of an old Category-5 Ethernet network, using half-duplex 10 Mbps NIC, is increasingly becoming something of an art form.
Do you utilise the host or LMHOSTS file in a bid to reduce NETBIOS name resolution traffic? Do you tweak your TCP/IP parameters, eg: increase the ARP timeout up from 2 minutes? You might use old bridge devices, in a bid to break up your collision domains - but still the broadcast domain persists! You make sure that, if you are using WINS, the new Burst Handling feature is on - and that DHCP is issuing an H-Node option tag to clients. You look forward to the procurement of a new 100 Mbps full-duplex fast-switched network.
Ethernet CSMA/CD has always been a play-off between contention versus ease of setup and function. You only have to look at Token Ring, as an alternative, to note the potential running cost increase - but do remember that Token Ring is a deterministic network: so that if you need to know exactly the worst- and best-case scenarios for guaranteeing that every client will transmit within a certain timeframe, then TR is for your business.
At some point you need to determine, at the very least, a data throughput baseline.
You should measure throughput after data transmission - because a system might be subject to delays caused by processor limitations, network congestion, buffering inefficiencies, transmission errors, traffic loads, congestion or inadequate hardware design. Throughput varies over time with traffic and congestion.
One way to calculate the actual data throughput is to use this formula:
Actual data throughput equals net utilisation, multiplied by efficiency rating, multiplied by wire speed.
Net utilisation is calculated by subtracting the number of collisions from the utilisation features.
The efficient rating is determined using the following table:
| Frame size (bytes) | Data size (bytes) | Overhead | Maximum efficiency |
| 1,518 (max) | 1492 | 2.5% | 97.5% |
| >1,000 | 974 | 3.8% | 96.2% |
| 500 | 474 | 7.4% | 92.6% |
| 64 (min) | 38 (no pad) | 50.0% | 50.0% |
| 64 (min) | 1(plus 37b pad) | 98.7% | 1.3% |
Example
Suppose that a network has the following characteristics:
a. A shared 10-Mbps Ethernet segment running at 40 percent utilisation.
b. 3% collisions.
c. Average frame size of 1,000 bytes.
The actual data throughput would be calculated as such:
Actual data throughput = (40 - 0.3) x 96.2 x 10 Mbps
Therefore actual data throughput = 3.56 Mbps.
There are of course many third-party tools available to measure network utilisation but the Microsoft Network Monitor which comes with Windows Server 2000 and Windows Server 2003 (although it is of course not the full version - you need to buy Microsoft SMS for that) is a very useful tool for packet analysis. It is equally useful for establishing what your data throughput parameters are at a given point in time (well between the measuring server and point of data origin).
The details panes are on the left side of the console; namely: network statistics, captured statistics, per-second statistics and network card statistics - all useful stuff.
These statistics windows will generate useful information of number of frames and type of frames (broadcast, multicast etc).
The progress bar in the right window will also record the high level mark advertising the max throughput strain at a given point.
Once you have a handle on data throughput you can decide if the network really is straining at the seems or just chugging along as well as your current setup allows; remember a standard Ethernet network is considered functional at only 30% throughput! In any event you will have proper records and statistics to put before the financial team in a bid to wrestle upgrade funds from them.
