OpenVPN CPU question

[R G G]

I have a 50/50 connection, when I use OpenVPN, that changes from 50/50 -> 20/50

I traced the problem to the CPU. For some reason, when it downloads, it only utilizes half of 1 core (I know OpenVPN is an inherently single-cored process). For upload, it uses all the core - which btw results in insane buffer bloat, but that's beside the point. My question is why does download use only half of 1 core's resources while upload uses the whole thing, clearly resulting in higher speeds. I know for a fact that it's not a network bottleneck since I receive 50/50 to the same server using OpenVPN's windows client.

If there is any script that I can run to force it to use all the resources it can, I'd appreciate if anyone lets me know.

Router model: RT-AC1750_B1 (RT-AC68U for Merlin firmware - latest update)

 

[R G G]

Can an admin move this to the Merlin forum, my bad.

Done. -rm

 

[skeal]

I have a 50/50 connection, when I use OpenVPN, that changes from 50/50 -> 20/50

I traced the problem to the CPU. For some reason, when it downloads, it only utilizes half of 1 core (I know OpenVPN is an inherently single-cored process). For upload, it uses all the core - which btw results in insane buffer bloat, but that's beside the point. My question is why does download use only half of 1 core's resources while upload uses the whole thing, clearly resulting in higher speeds. I know for a fact that it's not a network bottleneck since I receive 50/50 to the same server using OpenVPN's windows client.

If there is any script that I can run to force it to use all the resources it can, I'd appreciate if anyone lets me know.

Router model: RT-AC1750_B1 (RT-AC68U for Merlin firmware - latest update)

This model is not supported currently. Check this thread please.

 

[RMerlin]

This model is not supported currently. Check this thread please.

It probably works. I simply never added it to the list because I gave up trying to keep track of every single SKU variants Asus has for some of these models...

 

[skeal]

I have a 50/50 connection, when I use OpenVPN, that changes from 50/50 -> 20/50

I traced the problem to the CPU. For some reason, when it downloads, it only utilizes half of 1 core (I know OpenVPN is an inherently single-cored process). For upload, it uses all the core - which btw results in insane buffer bloat, but that's beside the point. My question is why does download use only half of 1 core's resources while upload uses the whole thing, clearly resulting in higher speeds. I know for a fact that it's not a network bottleneck since I receive 50/50 to the same server using OpenVPN's windows client.

If there is any script that I can run to force it to use all the resources it can, I'd appreciate if anyone lets me know.

Router model: RT-AC1750_B1 (RT-AC68U for Merlin firmware - latest update)

When you flashed to Merlin did you reset to defaults and manually reconfigure without importing saved settings in any way?

 

[L&LD]

I have a 50/50 connection, when I use OpenVPN, that changes from 50/50 -> 20/50

I traced the problem to the CPU. For some reason, when it downloads, it only utilizes half of 1 core (I know OpenVPN is an inherently single-cored process). For upload, it uses all the core - which btw results in insane buffer bloat, but that's beside the point. My question is why does download use only half of 1 core's resources while upload uses the whole thing, clearly resulting in higher speeds. I know for a fact that it's not a network bottleneck since I receive 50/50 to the same server using OpenVPN's windows client.

If there is any script that I can run to force it to use all the resources it can, I'd appreciate if anyone lets me know.

Router model: RT-AC1750_B1 (RT-AC68U for Merlin firmware - latest update)

I would guess the speed drop is due to where you're connecting. Does that endpoint also have a 50/50 up/down Mbps stable connection speed at a minimum as well?

 

[R G G]

I would guess the speed drop is due to where you're connecting. Does that endpoint also have a 50/50 up/down Mbps stable connection speed at a minimum as well?

I'm 1000% sure of the server, it's definitely capable of 50/50 and much more. It's my own VPS, I know what it's link speed is, it's around ~1.3 Gbps symmetrical. But I've even tried using several NordVPN servers. They all return the same result: 20/50. I also mentioned in OP that the OpenVPN client on PC connects with full speed, so it clearly has to do with the router.

When you flashed to Merlin did you reset to defaults and manually reconfigure without importing saved settings in any way?

Yes, in fact I've factory reset a couple times while on Merlin. Only remnant was traffic usage data. Have had this issue since 2 years ago, so it's not a problem with this specific Merlin release either.

 

[L&LD]

I'm 1000% sure of the server, it's definitely capable of 50/50 and much more. It's my own VPS, I know what it's link speed is, it's around ~1.3 Gbps symmetrical. But I've even tried using several NordVPN servers. They all return the same result: 20/50. I also mentioned in OP that the OpenVPN client on PC connects with full speed, so it clearly has to do with the router.




Yes, in fact I've factory reset a couple times while on Merlin. Only remnant was traffic usage data. Have had this issue since 2 years ago, so it's not a problem with this specific Merlin release either.

Have you tried different clients? Such as 1, 2 or 3? This will put the load onto a different core. This may be all that is needed.

https://www.snbforums.com/threads/openvpn-performance-of-the-rt-ac86u.41217/page-3#post-355706


If you've only reset a couple of times and this issue is outstanding for 2 years, I would recommend following something like the following to get your router to a good/known state.

https://www.snbforums.com/threads/rt-ac86u-factory-reset.56972/#post-496275

 

[R G G]

Have you tried different clients? Such as 1, 2 or 3? This will put the load onto a different core. This may be all that is needed.

https://www.snbforums.com/threads/openvpn-performance-of-the-rt-ac86u.41217/page-3#post-355706

I'll try all 5 clients and post the results. If that didn't work I'll try the configs you suggested.

 

[L&LD]

I'll try all 5 clients and post the results. If that didn't work I'll try the configs you suggested.

Shouldn't need to try all 5. Just odd or even clients will give you an idea if this is a possible cause.

 

[R G G]

Too late But this confuses me even more now.

Edit: Speedtest was a bit inaccurate, so I tried using a download accelerator. Results are a bit more clear now.

Core 1, 3, 5: 4.0 MB/s
Cores 2, 4: 3.0 MB/s

I don't have a consistent way of testing upload so I guess we'll have to go off the speed test numbers.

 

[RMerlin]

Core 1, 3, 5: 4.0 MB/s
Cores 2, 4: 3.0 MB/s

Which is normal. The router will always use core 0 to handle routing traffic, so I force the first client onto core 1 to spread the load. Client 2 will go to core 0, client 3 back on core 1, and so on.

On quad-core routers like the RT-AX88U, then a similar logic gets applied: Client 1 uses core 1, client 2 on core 2, client 3 on core 3, client 4 on core 0, and client 5 on core 1.

 

[R G G]

Which is normal. The router will always use core 0 to handle routing traffic, so I force the first client onto core 1 to spread the load. Client 2 will go to core 0, client 3 back on core 1, and so on.

On quad-core routers like the RT-AX88U, then a similar logic gets applied: Client 1 uses core 1, client 2 on core 2, client 3 on core 3, client 4 on core 0, and client 5 on core 1.

Sorry if I'm being slow here, but what does that have to do with the discrepancy in upload and download speeds? I've tested this router on a 1.5G/1G Bell FTTH connection as well, it gave me the same results.

 

[RMerlin]

Sorry if I'm being slow here, but what does that have to do with the discrepancy in upload and download speeds?

Nothing. Never said it did, I specifically quoted and answered why the speed was different between client1 and client 2.

 

[cplay]

Which is normal. The router will always use core 0 to handle routing traffic, so I force the first client onto core 1 to spread the load. Client 2 will go to core 0, client 3 back on core 1, and so on.

On quad-core routers like the RT-AX88U, then a similar logic gets applied: Client 1 uses core 1, client 2 on core 2, client 3 on core 3, client 4 on core 0, and client 5 on core 1.

In regards to openvpn on a ac86u based on what you said this would be the default layout of core usage:

Core 1 = cpu0 and by default all routing traffic on wireless and lan is on this core.

core 2 = cpu1 and vpn client 1, 3, 5 use this core

core 1 = cpu0 and is used by client 2, and 4

so for optimal speed on openvpn and if you are using just one vpn client then you should use clie t 1 that uses cpu1 and all wireless and Ethernet routing is handled by cpu0, correct?

 

[RMerlin]

so for optimal speed on openvpn and if you are using just one vpn client then you should use clie t 1 that uses cpu1 and all wireless and Ethernet routing is handled by cpu0, correct?

Routing is handled by core 0. No idea about wifi or NAT. You will have to monitor using top. On the RT-AX56U for instance, part of the routing/NAT acceleration is handled by core 3.

 

[cplay]

Routing is handled by core 0. No idea about wifi or NAT. You will have to monitor using top. On the RT-AX56U for instance, part of the routing/NAT acceleration is handled by core 3.

cheers, so client 1 using cpu1 which is actually using core 2, right?

as you’ve explained routing is done via cpu0 which is core 1, routing refers to handling the connection, processing requests etc, lan?

 

[RMerlin]

cheers, so client 1 using cpu1 which is actually using core 2, right?

Yes. Clients are shifted by one.

 

[cplay]

Yes. Clients are shifted by one.

Cheers, is it the same layout with routing on cpu0 and vpn client 1 on cpu1, vpn client 2 on cpu0 if using stock?

 

[RMerlin]

Cheers, is it the same layout with routing on cpu0 and vpn client 1 on cpu1, vpn client 2 on cpu0 if using stock?

I don't know. All I can tell is I've had discussions with Asus in the past about how to best optimize things, however their OpenVPN code is now closed source, so I have no idea what/how they are doing anything.