It has been confirmed to me that the ER-X / ER-X-SFP have only one 1 Gb/s link between the SoC and switch. Since every packet that enters the SoC through that link will have to exit the same path, 1 Gb/s is the maximum aggregate throughput.
I’ve not been able to garner any interest in why bi-directional testing take a substantial performance hit. I may try some earlier firmware releases in the future but for now I’m moving on from this subject.
See Preliminary Observations of the ER-X for the story leading up to now. I retrieved the ER-X-SFP but after giving it some more thought I concluded that the SFP port shouldn’t provide better throughput because it has to be hanging off the switch.
But I decided to test away anyways. Here’s a baseline iperf run of my test setup:
About what I’d expect from the hardware I’m using: 908 Mb/s uni-directional, 1,465 Mb/s aggregate bi-directional.
Here is the ER-X, with eth0 WAN and switch0 / eth1LAN:
The bi-directional result of 765 Mb/s is one of its better runs, high 600s to low 700s was its general range.
Now, here’s the ER-X-SFP in the same config:
It takes a hit, tho it’s results were consistently in the high 700s — a tiny bit better than the ER-X.
Now let’s try it again using an RJ45 SFP in eth5 as WAN:
929Mb/s bi-directional! And consistent! So it performs better… and I’ve no idea why… but clearly the platform is limited to 1Gb/s aggregate throughput.
As a sanity check I ran the same test against an ERLite-3 and was able to get > 1,700 Mb/s using multiple threads. Using multiple threads against the ER-X did not affect the results.
Both ER-X and ER-X-SFP on v1.9.1.1, configured using the Basic Setup Wizard for single LAN, with set system offload hwnat enable and port forwards for 5001, 5201, and 5202 TCP & UDP (iperf / iperf3).
I’m going to work up a thorough review soon, but preliminary testing confirms what others have seen: Bi-directional performance of the ER-X is sub-1Gb/s aggregate and a fair bit lower than uni-directional performance. I’m seeing over 900 Mb/s uni and low 700 Mb/s for bi-directional NAT’d traffic on v1.9.1.1.
I’ll do more testing with routed non-NAT traffic next week.
I’m surprised more attention hasn’t been brought to this given how often the ER-X is promoted as a cheap router for Google Fiber and other Gigabit FTTH offerings. The reality of the ER-X is that it’s more like a 350/350 router than Gigabit.
In trying to understand why this is, I came across a blog entry speculating that one of the 1Gb/s interfaces from the SoC is reserved for the SFP on the ER-X-SFP. I’ll be swapping one of these ER-X’s for the ER-X-SFP at my girlfriend’s home to test if there’s a performance improvement using eth0 and eth5 on the ER-X-SFP vs eth0 and eth1 on the ER-X.
Another item of note: Pass-through PoE does not require a 24v PSU! The included 12v PSU is only 0.5A and not suited to powering anything else — it will, but expect stability problems under load, or perhaps a fire — but any 9v-26v PSU that supplies more watts will work. An upgraded 12v PSU will generally be a few bucks cheaper than 24v.