Difference between revisions of "Hardware: ZyXEL PLA-4201 Twin Pack - throughput test"
Lukas Dzunko (talk | contribs) (→Methodology of synthetic tests =) |
Lukas Dzunko (talk | contribs) (→Methodology of synthetic tests) |
||
| Line 11: | Line 11: | ||
I prepared several tests. Here is description of tests: | I prepared several tests. Here is description of tests: | ||
== Methodology of synthetic tests == | == Methodology of synthetic tests == | ||
| + | First of all i waited for everything to settle down and then at each test point I executed following tests. I run first test once mode if it result in high response rate. It looks like that ZyXEL powerline adapters needs some traffic to "sync" and configure connection. | ||
| + | |||
=== Test #1 === | === Test #1 === | ||
| + | <pre>ping -c 10 server</pre> | ||
| + | Simple ping test to see reaction of connection on as little traffic as possible (no other communication was flowing thought extenders during this and also other tests) | ||
| + | |||
| + | === Test #2 === | ||
| + | <pre>ping -c 100000 -f server</pre> | ||
| + | Ping food test with default ping size. This test is generating relatively small packets at speed of source network card. Purpose of this test is to see packet lost and response time if queuing occur. | ||
| + | |||
| + | === Test #3 === | ||
| + | <pre>ping -c 100000 -f -s 1400 server</pre> | ||
| + | Same as Test #3 but with size close to MTU. I selected smaller size as devices from some vendors have great throughput at size of MTU but can't maintain it when packed size is smaller. | ||
| + | |||
| + | === Test #4 === | ||
| + | <pre>netperf -l 60 -H host</pre> | ||
| + | "One minute" classic netperf test to see maximu throughput. | ||
| + | |||
| + | === Test #5 === | ||
| + | <pre> netperf -t omni -j -l 60 -H host -- -d rr</pre> | ||
| + | Omni netperf test mainly to see number of transactions but other results of this tests are also interesting. | ||
| + | |||
| + | === Test #6 === | ||
| + | <pre> netperf -t omni -j -l 60 -H host -- -d rr -O \ | ||
| + | "MIN_LATENCY,MAX_LATENCY,P50_LATENCY,P90_LATENCY,P99_LATENCY,MEAN_LATENCY,STDDEV_LATENCY"</pre> | ||
| + | Same omni test as in #5 but configured to show latency. I expect that test #2 and #3 show min/avg/max but it is also interesting to see this via percentile calculation. (e.g. several delayed packets can affect average calculation but they will be omitted in percentile output) | ||
Revision as of 23:50, 19 March 2013
It is not so easy to add wires/connection once flat or house is build. Even when people think about devices placement they can end up with missing Ethernet connection for some device. WiFi connection is not usable for all kind of network communication due to nature of it. WiFi have limited throughput, jitter and other kind of problems caused by "wireless" connection. So is powerline connection solution to this problem ? At beginning of this document i can say: "I don't know", but i try to find answer.
I bought ZyXEL PLA-4201 recently and I use this device during tests to find answers. I know how this device work by theory but I don't know what is in real word. In theory it should modulate all data to mains with hope that second device "catch" those data without too much distortion but there are several problems:
- switching power supplies and power regulators (light dimmer, vacuum cleaner "speed" settings) are creating serious noise covering wide range of frequencies
- power cables are not optimized to transfer "data". Most of the filters kill anything else that 50/60 Hz.
- if sockets are not on same power line then communication (as expected) is not possible
- communication can't pass transformers, surge protectors and other similar devices
Power cables are definitely not "polite" place for Ethernet data, but why don't test it ? Right ? In office I heard opinions like: "it will work only on 10 meter cable or shorter", "noise kill all transferred data", etc. Will it be true ?
I prepared several tests. Here is description of tests:
Contents
Methodology of synthetic tests
First of all i waited for everything to settle down and then at each test point I executed following tests. I run first test once mode if it result in high response rate. It looks like that ZyXEL powerline adapters needs some traffic to "sync" and configure connection.
Test #1
ping -c 10 server
Simple ping test to see reaction of connection on as little traffic as possible (no other communication was flowing thought extenders during this and also other tests)
Test #2
ping -c 100000 -f server
Ping food test with default ping size. This test is generating relatively small packets at speed of source network card. Purpose of this test is to see packet lost and response time if queuing occur.
Test #3
ping -c 100000 -f -s 1400 server
Same as Test #3 but with size close to MTU. I selected smaller size as devices from some vendors have great throughput at size of MTU but can't maintain it when packed size is smaller.
Test #4
netperf -l 60 -H host
"One minute" classic netperf test to see maximu throughput.
Test #5
netperf -t omni -j -l 60 -H host -- -d rr
Omni netperf test mainly to see number of transactions but other results of this tests are also interesting.
Test #6
netperf -t omni -j -l 60 -H host -- -d rr -O \ "MIN_LATENCY,MAX_LATENCY,P50_LATENCY,P90_LATENCY,P99_LATENCY,MEAN_LATENCY,STDDEV_LATENCY"
Same omni test as in #5 but configured to show latency. I expect that test #2 and #3 show min/avg/max but it is also interesting to see this via percentile calculation. (e.g. several delayed packets can affect average calculation but they will be omitted in percentile output)
