Actiontec ECB6200 MoCA Adapters iperf3 Results. Download Speed is Lower than Upload?

[bmn1]

I've used powerline adapters for a few installs in the past which work well but don't really provide a lot of throughput. So I decided to pick up a set of the Actiontec ECB6200 MoCA adapters just to check them out . So far I'm only "half" impressed with the results...

I purchased these brand new a few days ago. The firmware version on them is 2.11.1.50.6200.7

To keep things simple, and to provide what I would consider "ideal conditions" I decided to connect both of the ECB6200 Adapters to a 60' spool of RG6 Coax. Then one adapter is plugged into the LAN port of my laptop via 3' of CAT6 cable, the other is plugged directly into the LAN port of my router via 3' of CAT6 cable. My desktop is also plugged directly into my router via 3' of CAT6 cable as well.

Testing was done using the latest version of iPerf3 v 3.1.3. Here is the command line syntax I used for testing...

iPerf3.exe -c 192.168.1.100 -t 10 -w 1M -O 3

-t: 10 second duration
-w: 1MB Windows Size
-O 3: Omit the first 3 seconds of testing (to allow a warm-up)

As you can see, the upload speeds look amazing! However, the download speeds are about 2.6x LESS than the upload speeds! If I plug my laptop directly into my router (bypassing the ECB6200 adapters) and the run the test again, I get a solid 970+ mbps Upload AND Download speeds. So obviously that only leaves the ECB6200 adapters to blame.

I also read a few reviews from people saying that they noticed a decrease in transfer speeds after the ECB6200 adapters have been plugged in for awhile. So I decided to leave them plugged in and connected for about 30 minutes and then retested them again. The results were "Upload: 513 mbps / Download: 335 mbps". So this time my upload speeds were HALF of what they were when I ran the first test when they had only been plugged in for a few minutes. I tested them again three more times and the results were the same.

Then I decided to unplug both ECB6200 adapters and leave them sit for about 15 minutes and then I tested them again. The results were "Upload: 950 mbps / Download: 364 mbps", basically the same as my first test. So clearly the ECB6200 adapters suffer from heat soak which is probably why Actiontec attempted to combat this by releasing a revised version of the ECB6200 which has vented holes in the casing that the original version didn't have.

I'd say these units have some awesome potential, even at 365 mbps, they still outperform any powerline adapter that I've tested. But that upload/download split is definitely disappointing. Especially since it seems like the units are obviously capable of near Gigabit Upload speeds, yet only 365 mbps Download speeds.

Has anyone else has experienced either of these issues as well?

 

[degrub]

i think there were some options that needed to be set with iperf mentioned in one of the threads in here. They saw similar issues with the testing. Also, there was a way to access the diagnostic page of the adapters to check the physical rate the modems are syncing at.

 

[bmn1]

Yeah but I don't see any of the iPerf parameters being the issue seeing as how I was able to get 950+ mbps Upload. I ran the test exactly the same way just in Download (reverse) mode. iPerf sent the packets the same exact way for the Download test as it did for the Upload test.

As for the diagnostic page, yes I already checked them, it said they were both linked at 1 Gpbs.

 

[krkaufman]

As for the diagnostic page, yes I already checked them, it said they were both linked at 1 Gpbs.

That sounds like an Ethernet link rate. The poster was suggesting checking the MoCA PHY rate (likely 600+ Mbps) and whether the adapters were using two bonded channels. (Answer appears obvious, since 500 Mbps is the max for a single channel.)

As for iPerf, look for a “threads” option. Looks like it may be -P. Try the test with an increasing number of parallel connections.*

* I’ve no clue how a different result using parallel threads might translate to real world performance.

 

[umarmung]

1. Check the results by actually reversing which is the server and which the client at least once, i.e. not relying on the reverse mode command line option working as you may expect.

2. It is recommended to use the "-P" option, "-P 10" is fairly standard. There used to be significant variation between using and not using this parallel streams option, especially on platforms with weaker CPUs and Gigabit throughput.

3. Excluding the reverse mode option, it is the server report that matters, not the client, especially if you use the UDP options (not being used here and should not be used if only testing bandwidth).

Finally, I am not sure if you a professional installer or home user when you refer to installs, but in general, powerline is an inferior connectivity option to both Ethernet and MOCA for multiple reasons. It is often inferior to even WiFi. In particular, you will be lucky to get a fraction of the test speeds reported here in an actual install. If you already know that or you are purely relative testing between powerline adapter products, then this is for the benefit of others reading the thread.

 

[bmn1]

1. I already tried that during my original testing which got me the same exact results from both machines. The Upload being ~350 mbps and the Download being ~950 mbps.

2. I retested using the "-P 10" switch which did net me ~940 mbps in both directions, however, it makes absolutely no sense to test using parallel streams when the majority of tasks and applications aren't going to be using parallel streams when uploading/downloading. I'm more interested in real world results, not the fastest number I can obtain via a synthetic benchmark.

3. Server window shows the same results that the Client window shows.

I do computer work for a living but I also do Surveillance Camera and Home Automation installs on the side. Yes I'm well aware how inferior Powerline Adapters are, however, I have come across far too many situations where a Powerline Adapter was the more ideal solution over other options. Ideally, I would love to just run an ethernet cable for everything but that's hardly a practical solution for a lot of situations. I'm not sure which Powerline Adapters you've tested but I have used the Netgear Powerline 2000 adapters for a few installs and they have been working great so far. I did some similar testing with the Powerline adapters like I did with the Actiontec adapters. Below is some of the results I obtained.

In a 3 story house, built in 2015.
- When testing with two powerline adapters in the same room: Upload: 485 mbps / Download: 200 mbps
- When testing with one on the 2nd floor and one on the 1st floor: Upload: 90 mbps / Download: 130 mbps
- When testing with one on the 2nd floor and one on the 3rd floor: Upload: 92 mbps / Download: 75 mbps

Obviously I don't think they are the greatest solution out there, hence the reason for picking up a set of MoCA adapters to do some testing.

In any case, back on topic. I checked the Node Info page on each of the adapters to get the Physical Rate as another member suggested, and here is what it's showing for both nodes...

Now if the "Phy Rate Mbps" is only showing "670 mbps" then why am I able to get ~950 mbps with the Upload test? Shouldn't I be maxed out at 670 mbps? And why the hell can I get ~950 mbps while sending (Upload) but not receiving (Download)? It makes no sense. =/

 

[krkaufman]

2. I retested using the "-P 10" switch which did net me ~940 mbps in both directions, however, it makes absolutely no sense to test using parallel streams when the majority of tasks and applications aren't going to be using parallel streams when uploading/downloading. I'm more interested in real world results, not the fastest number I can obtain via a synthetic benchmark.

As I mentioned earlier, I don’t know why the -P option makes a difference, but stating that not using it produces results closer to “real world” results would also appear to be an assumption.

I checked the Node Info page on each of the adapters to get the Physical Rate as another member suggested, and here is what it's showing for both nodes...

Now if the "Phy Rate Mbps" is only showing "670 mbps" then why am I able to get ~950 mbps with the Upload test? Shouldn't I be maxed out at 670 mbps? And why the hell can I get ~950 mbps while sending (Upload) but not receiving (Download)? It makes no sense. =/

Looking at the graphs, the adapters appear to be in bonded* mode, so you’re using 2 MoCA channels to communicate between the adapters, so the effective PHY Rate is doubled ... 670x2 = 1340 Mbps. And the TX power levels look great, as well, at -27 dBm (min power is -30, and I’ve never seen it).

* edit: Being the only 2 MoCA nodes on the coax, they actually appear to be in TURBO bonded mode, where they still have the same PHY Rate but some overhead is avoided netting a bump in effective throughput (1 Gbps vs 800 Mbps).

 

[bmn1]

As I mentioned earlier, I don’t know why the -P option makes a difference, but stating that not using it produces results closer to “real world” results would also appear to be an assumption.

I wasn't asking "why" the -P switch makes a difference, I already know why it makes a difference, because it uses multiple streams instead of a single stream. Same reason why download managers can usually download from a server faster than standard web browsers, because they are using multiple connections to the server rather than a single connection.

Using the "-P" switch in iPerf is great for stress testing a network, but hardly useful for wanting to obtain real world speed test results since most applications are only going to use a single stream/connection rather than multiple. But hey, if people want to go around cheating the test results just to make themselves feel better, then by all means. I'd rather continue to look for a real world solution.

 

[Samir]

Late to the party, but your results are interesting. I have a pair of g1100 frontier routers that I've tested as a moca link in a similar scenario as yours and the physical link they showed was 600Mbps with a longer run and 630Mbps with a shorter run. I ran an iperf test on the 600Mbps link and it was almost 500Mbps, but I may have used the -P switch.

As far as parallel streams, many, many applications use it and the difference is noticeable when you have more streams available. I have an old example, but it still holds water.

Back in the mid-2000s, bandwidth everywhere was much less than today and a 8M/256k service was the fastest I could get. But I needed more upload bandwidth--much more. So I got 3 cable modems, each with an 8M/256k service and used a Cisco rv016 multi-wan router to put them all on the same network.

Now, the way the rv016 works is very similar to how many multi-wan routers work--they don't actually bond the different wan connections; rather, they send various requests from the lan to the least used wan connection or using some other type of load balancing like a round-robin algorithm. In other words, a single stream could only use up to 8M download or 256k upload no matter what, but that multiple streams would increase the overall amount of data being sent.

So for uploading the large files I was creating at the time, this meant using a few computers to saturate the upload--no big deal. But what I didn't realize until I noticed much snappier web page loads (24Mbps was almost unheard of at that time), was that web pages consist of up to hundreds of little streams--each image, or piece of code, or box of text. So when I started looking at how web pages were loading, the page elements were loading randomly all at once since all the modems were being used since there were so many streams.

Fast forward to today and everything is bonding and streams. Cable modems are multiple bonded channels, the moca adapters are using bonded channels, even the cpus in computers use multiple channels for the memory and multiple cores for processing. So having multiple streams is a real-world thing.

And final case in point is the testing I've done on some ipsec vpn connections I have that connect various locations around the country. A single stream on iperf will show as little as 2Mbps on a 10Mbps link. But the real work usage shows a lot closer to 10Mbps when it is multi-stream. And an iperf test with just 3 streams maxes it out to the full 10Mbps.

Some food for thought.

As far as your testing, the oddball speeds also sound like an Ethernet duplex mis-match. And this one is easy to test for--swap the computers on each of the link. If the slow speeds follow a computer (move from one moca adapter to the other), it could be a duplex mis-match with that computer. If you've already done testing both ways with both systems (which I think you have), try manually setting the duplex on the computer if it lets you, or attach an adapter to a switch which the computer connects to. I've seen some weird stuff happen when Ethernet negotiation borks. Good hunting and I'd love to know what else you've figured out about these as I have a pair on the way that I plan to use with a gigabit Internet plan, so I'll need all the bandwidth that it can provide.

 

[krkaufman]

I ran an iperf test on the 600Mbps link and it was almost 500Mbps,

Which would be in-line with a TURBO standard MoCA 2.0 connection, possible where just 2 MoCA nodes are present ... and the G1100 being standard MoCA 2.0.

 

[Samir]

Which would be in-line with a TURBO standard MoCA 2.0 connection, possible where just 2 MoCA nodes are present ... and the G1100 being standard MoCA 2.0.

I tried to research what the turbo standard is for 2.0, but couldn't find anything. Can you post a link?

 

[Swistheater]

what is the difference between D-low band, D-high band, and D -extended band- does picking one over the other guarantee better performance, also which is better when in combination with a cable network in the house.