OpenVPN performance of the RT-AC86U

[L&LD]

The server is directly connected to the openvpn at my place. The server only has an i3 processor but I think it should be enough for higher speeds than what I'm getting.

The i3 should be powerful enough. However, is it actually running the OpenVPN client itself?

 

[mafiaboy01]

The i3 should be powerful enough. However, is it actually running the OpenVPN client itself?

The openvpn server is run on the ac86u and the openvpn client is run on the i3 server.

 

[Xentrk]

I used ipvanish in this test.
fw: 384.9

Very nice video. Thanks for sharing.

I recommended an AC86U to a friend based on the improved OpenVPN performance others have reported in this forum. His use case is to stream media to circumvent geo restrictions. Unfortunately, I was never able to achieve the improved OpenVPN performance others reported. It was no different than my AC88U.

Very happy with the OpenVPN performance on my pfSense appliance. I converted an old Windows 7 PC I to a pfSense appliance with an i5 using AES-NI.

 

[L&LD]

The openvpn server is run on the ac86u and the openvpn client is run on the i3 server.

Wired or wireless connection on the i3 server?

 

[mafiaboy01]

Wired or wireless connection on the i3 server?

Everything is wired. Both the computer on my end(i7) and the server are wired.

 

[Odkrys]

proto udp

try on tcp.

 

[RMerlin]

And also add this to the custom settings:

txqueuelen 1000

 

[mafiaboy01]

Thanks! Changed to TCP and set the txqueuelen 1000 setting on the router.

Now averaging between 100-200Mbps! Backing up my computers will be so much faster now.

 

[Vince]

And also add this to the custom settings:

txqueuelen 1000

can vpnclient set this value forcing a VPN-service, or must it be set manually on both sides?

 

[RMerlin]

can vpnclient set this value forcing a VPN-service, or must it be set manually on both sides?

That queue len is related to the local network interface. So if you are using two Asus routers, then both of these default to 100, and would benefit from increasing to 1000. If you are connecting to an endpoint not using Asuswrt, then it depends on what they use within their own configuration.

I know other router manufacturers also suffer from a similar issue (defaulting to 100 on their TUN interface).

 

[Vince]

Very happy with my RT-AC86U, Merlin 384.10_2 and NordVPN on TCP, there is only 1 thing troubling me sometimes.
The average loss in speed compared to no VPN is only 4%
But sometimes NordVPN server active on my system, drops its performance to say 50% and I have to change to another one to regain top-speed.
Currently I have 3 servers defined in "client instance" slot 1, 3 and 5 and using 1 of those 3
Would it be feasible to add a "fastest" option in "client instance" which option when selected would test performance (maybe through ping test) of the 5 possible entries on a regulair basis say once per hour and close active VPN-server and open the fastest found or only switch to the fastest when it can be improved by for example 10%?

 

[kamoj]

Easier said than done...
https://www.snbforums.com/threads/h...lient-using-voxel-firmware.55646/#post-473243
https://www.snbforums.com/threads/n...art-vpn-server-from-gui-on-r7800-voxel.55844/

 

[RMerlin]

If performance is your focus, then avoid VPNs. You are basically trying to shovel all your traffic through the same pipe used by all other users of the same VPN server. Those servers only have a fraction of the bandwidth that your ISP can provide. Congestion *will* occur at one point or another.

 

[Vince]

Well, its not that bad, pay my ISP for 100Mbps and in 90% of the time I get 96Mbps over NordVPN tcp connection at 1 selected server.
If this server congests and drops to say 50%, another of my 3 defined servers delivers 96Mbps at the same time.
So selecting from defined servers the fastest VPN server on the fly by clients router would I think be a unique very appreciated feature by VPN users.
Just have no idea how difficult/impossible/time-consuming, implementing such a feature is

 

[kamoj]

One way to do it could be to change server if current server is not among the 5 "least loaded" ones,
and not change too often, and current load is too low, or load difference is bigger than x % etc...
That doesn't need a lot of code to accomplish.

I'm considering it for the Netgear R7800 Voxel FW / Kamoj Add-on.
(To ping the servers only results in the physically closest server.)

You can get server with lowest load this way:

curl --silent 'https://nordvpn.com/wp-admin/admin-ajax.php?action=servers_recommendations' | jq --raw-output 'limit(1;.[])|.hostname'

Now to the problem - again:
But in my tests, the server with lowest load (according to NordVPN) is not necessary the fastest one.

 

[Vince]

Thanks kamoj,
same observations here, NordVPN's data of recommended servers does not help, often they are very slow.
The best way getting the fastest server I found was ping-testing, used "IP tools for excel" for that, I see that such a nearby server can have a high load, but still for me they where mostly the fastest. Don't know if ping result could indicate if such a server gets high loaded though.

 

[DigitizedMe]

I'm interested to see if anyone can run a benchmark between LZ4 and LZ4v2 compression for science! I'm limited by bandwidth with 100mbps down and 10 up. (I'm interested for future use.)

Eric said the bottleneck isn't in the cipher regarding AES-128-GCM vs AES-256-GCM. I'm wondering how much the modification to the compression algorithm actually improved performance.

*I searched for "v2" on each page of this thread since the search term was too small. I didn't find anything.

 

[doczenith1]

I'd be willing to perform some science for you when I have some free time over the next few days. That said, am I remembering correctly in that the server controls the level of or use of compression regardless of what the client request? Or am I getting options mixed up and it was some other option that behaved that way?

 

[The Chief]

RT-AC86U CHACHA20-POLY1305 benchmark.

So truch is that there is no gain from CHACHA20-POLY1305 cipher on AES-accelerated devices, like AC86U?

 

[Odkrys]

So truch is that there is no gain from CHACHA20-POLY1305 cipher on AES-accelerated devices, like AC86U?

Yes