Hello! I use a mesh (802.11s) network of three WAX202 nodes. I employ batman-adv for distributing multiple VLANs over the mesh network.
Now, I have two VLANs 10 and 20. VLAN 10 is allowed to be forwarded to 20. I test routing perforance by accessing VLAN 20 from VLAN 10, and then by running a test only within VLAN 20.
- Two PCs, each one connected to one access point via cable
- The access point at PC1 is the router, the other is just an access point without routing
- Wireless link between access points
- Two test cases (A and B) with iperf3
PC1 on VLAN 10 ---> PC2 on VLAN 20, gives around 125 MBit/s
Then, I configure the network port on the first access point to VLAN20 instead. So, both machines are on the same VLAN.
PC1 on VLAN 20 ---> PC2 on VLAN 20, gives around 220 MBit/s
Why is it so much faster? Does the routing performance of the WAX202 hurt speeds that much? I saw around 50% idle on the router for test A.
Thanks for the advice. I was not aware of these options before.
Packet steering was on. Switching it off does not change performance, so I left it on.
Software/hardware offloading was off. Switching it on does improve performance slightly. I am now at a around 147 MBit/s between VLANs (test B).
Is it expected to have such a loss in performance when routing betwenn zones?
Tbh, I'd have to test in an environment. Although, I get 850 Mbps plus on lan-to-wan. I don't beleive lan-to-lan should be any different.
I will do a cable test then, just to be sure.
Ok, the cable test was conducted as follows: Two PCs on two different LAN ports of the router.
VLAN 10 --> VLAN 20
Vlan 10 --> VLAN 10
So, no loss when routing between VLANs/zones over LAN. I am not sure what it could be then. It only degrades performance when I traverse my wifi mesh.
This is just a wild guess, but could this be related to a difference in download versus upload throughput over WiFi? You could do some iperf3 tests using a single WAX202 device in a simple non-mesh configuration to see if there is a performance difference in this regard for the wireless hardware.
Explain how they're connected, not just the VLANs. How many devices involved...how many SSIDs?
If you're on the same WiFi and you're using 2 different SSIDs on the same band for example, 50% is expected (in theory).
Expanding on the opening post:
- I use three WAX202 in an 802.11s mesh (5 GHz only), one of which is also the router
- Then, I use batman-adv for mesh routing and VLAN tunneling
- On each device, I use a single bridge for bridge VLAN filtering
- On each device, there are interfaces that each attach to one of the VLANs (e.g. lan and guest)
- There are two 5 GHz WIFI networks (AP) on each node (for lan and guest networks), connected to the aforementioned interfaces
- mgmt (10)
- lan (20)
- guest (30)
- iot (40)
- vpn (90)
I never change anything regarding the wifi configuration between tests. I only connect the wired host on one of the access points to a different VLAN, by changing the setting in the access point for the LAN port.
It's still not clear to me where test client and test server are located - and how they're connected.
Nonthless, if wired bandwidth is around 900, you seem okay.
This is how it is connected in my wireless use case (where I have performance issues)
For test A (slow), the two measurement hosts are connected to different VLANs (e.g. 10 and 20).
For test B (fast), the hosts are both on the same VLAN (does not matter which one).
The wired case that I used for testing looked like this. Here, I have no performance issues.
(Sorry if this is frustrating)
It really seems normal.
- No info about the established WiFi link, what it established rate streams, etc.
*WiFi availability/congestion on channel
- Just to be sure, you're inquiring about the bandwidth still? (The VLAN thing, not sure why that's concerning you.)
- To be clear, you said same speed testing on both VLAN 10 or 20, correct?
My issue is that on the same wireless link and with the same PCs, connected to the same access points, I do not get the same results between these two cases:
- the two PCs are on the same VLAN
- the two PCs are not on the same VLAN
Having the PCs on the same/not the same VLAN is the only thing that changes between the tests. Having them on different VLANs costs me bandwidth vs. having them on the same VLAN, which I don't understand.
I didn't discuss the quality of the WIFI link itself, because it stays the same. The measurements are also reliably repeatable at any time. Anyway, I use channel 36 and the established rate to the AP with the other testing PC is like this:
There were no other networks on that channel at the time of testing.
Finally, why this is relevant to me: The VLANs offer different services and VLAN 10 has access to VLAN 20. From a machine on VLAN 10, I want to be able to use services on VLAN 20 with full speed.
If you're going to test inter-VLAN routing, you should do this connected by ethernet with no wireless links in-between. So make use 2 physical ethernet ports on your router into access ports -- one for VLAN 10 and the other for VLAN 20. Connect your computers to those ports and then run iPerf3 across them.
Right now, there are multiple conflating factors when you include wirelss in the mix.
I did such a cable test already (a few posts above) and there was no throughput penalty that way. Both variantes yielded 900 MBit/s in iperf3.
So the issue is related to wifi, not VLANs.
I guess that is the only valid conclusion. My measurements over wifi might not be so reliable.
Anyway, thanks to everyone for their help! I will just keep everything as is
Well, the measurements are data points that include many many factors -- from the link rate and signal quality between the different mesh points, to the link between the PC and the mesh point, to the tuning of the wifi radios, the pyhsical characteristics of the environment (construction materials, etc.) that can have an effect on the performance of wifi, the wifi radio capabilities in the AP, and also the capabilities in the PC.... the list goes on. You can try to troubleshoot/characterize those things, but it also depends how much effort you want to put into it and what practical improvements you think you can make.
If your problem is solved, please consider marking this topic as [Solved]. See How to mark a topic as [Solved] for a short how-to.
Yes, you are right. I might try and isolate aspects of my network and do some more benchmarking later.
This topic was automatically closed 10 days after the last reply. New replies are no longer allowed.