Understanding network test

[staypuppy]

I've read the manual and every relevant post I can find, but still cannot find an answer as to why the network test is not working for me.

When I run the standard test with the duty cycle at 100% and watch Networking in the Task Manager I never see the graph go over .5% (yes .5). When I use the advanced networking test I get better results (25-50%) but I never see anything close to 100%. We're looking for software that can keep the network pegged at close to 100% for a sustainable amount of time. Is this possible with your software?

I should note that the Advanced Test will not work for us because we need to run tests on a single isolated machine.

[David (PassMark)]

Using BurnInTest you probably aren't going to get to 100% utilization of your network bandwidth.

This can be for several reasons,

1) CPU bottlenecks and PCI bus bandwidth can restrict the amount of data sent & received. This is especially the case of GB networks.

2) With most protocols it is impossible to get data throughput that equals 100% of the theoretically capacity. This is because of protocol overheads and latency.

3) The standard network test is a ICMP ping style test (with some extra payload and check-sums). Doing a ping style test with the local host address (127.0.0.1) doesn't result in any physical data transmission on the Ethernet port for most hardware. (I assume you are using the address 127.0.0.1 on Ethernet?). So it is a poor test that doesn't fully test the hardware. A more effective test is to test against a remote machine & we recommend doing this.

4) The standard network test sends a packet, then waits for the response back from the remote machine. So there is a period without any data transmission while the test waits for the response and then a further small delay while the check-sum is calculated and the echoed data is verified to be correct.

5) The duty cycle setting in BurnInTest can throttle the test. (but you stated you had this set to 100%). Having other tests running at the same time can also reduce the throughput do to a lack of CPU time.

6) You might need to tweak some Windows TCP/IP settings, like the packet size and sliding window size to get the best performance. But the optimal settings depend on your type of network. (Windows tries to auto-config and does an OK job most of the time)

7) If you are using a wireless network then the manufacturer's speed claims seem to be vastly over rated.

To test the maximum throughput (without doing any error checking) I would suggest using our PerformanceTest product. On a 100Mbit/sec Ethernet network you should get to between 80% - 90% utilization. On a 1Gbit/sec Ethernet network you should get to between 70% - 80% utilization. Of course this depends on the factors stated above, like CPU power.

[staypuppy]

Thank you for taking the time to give such a detailed response. It is very much appreciated. I'll check out Performance Test. Overall I've been very happy with the trial testing against the CPU and the RAM. It was only the networking that was frustrating.

[pcsmo]

The wireless AP in the test wasn't using a "proper" channel (9). Only 1, 6, and 11 are the only channels that don't overlap. http://www.cisco.com/en/US/docs/wire...e/Channel.html

I agree with the overall point of the article though.