Gargoyle Performance

[pmerrill]

Some performance measures for my 100 Mbit link from Optus.

Gargoyle 1.4.6, TP Link 1043nd

If I connect directly to the cable modem (no router) I get 98 Mbits/sec via speedtest.net.

If I go through Gargoyle I get about 50 Mbits/sec (with QoS on). With QoS off I get about 75 MBits/sec.

As a comparison, stock performance on my TP-Link 1043 nd I get 67 Mbits with their QoS on and about 97 Mbits with it turned off.

Seems that stock is faster than 1.4.6 but with regard to wireless, wireless n stock crashes whereas Gargoyle has no problem.

Is this level of performance what should be expected?

Are there any tweaks that one can apply to improve the performance?

What about an upgrade to 1.5.3???

[ispyisail]

http://www.gargoyle-router.com/wiki/dok ... =swap_file

[pmerrill]

So what are you saying? If I upgrade to 1.5.3 and put a swap drive on the 1043nd with a USB key I'll get better performance?

I note from the status page that I have Memory Usage: 12.9MB / 28.8MB (45%) which seems to indicate I have free memory, so I'l not sure what the USB key will do???

Out of curiosity, what's the overall network performance difference between 1.4.6 and 1.5.3?

[mix]

A swap file will do nothing for wan performance. There should be negligible difference between network throughput between 1.4.6 and 1.5.3, assuming you aren't using QoS. QoS is an extra level of packet inspection; it will reduce your bandwidth and increases your cpu load (which will probably also decreases your throughput). I don't know how many of us have a 100 Mbps internet connection to do throughput testing on like you do. I don't even know if I would trust an internet speedtest at those speeds to be accurate. I am a little surprised you aren't getting the full 100 Mbps with an N router like your 1043nd though. You aren't testing the throughput over wireless right? I doubt you will ever get 100 Mbps with wireless.

[pmerrill]

No no, not over wireless, a wired 1 Gbit connection to the router.

Here's the speed test with QoS on

http://www.speedtest.net/result/1786578923.png

and here it is turned off

http://www.speedtest.net/result/1786583163.png

The performance is not up to the par of the stock firmware which was generally 97 Mbits/sec with no QoS. However, as I said, at least wireless n does not drop out!!

What about the % Bandwidth at Capacity? Does it make any difference what that setting is? Mine currently is:

Service Class Name Percent Bandwidth At Capacity Minimum Bandwidth Maximum Bandwidth Load (kbps)
Voip 1% 80 kbit/s unlimited 0.0
Fast 70% zero unlimited 0.0
Slow 1% zero unlimited 0.7
Normal 28% zero unlimited 0.0

[nicram]

Connect to the router with putty.
Do:

Code: Select all

cd /etc/init.d/
./bwmon_gargoyle stop
./bwmon_gargoyle disable

It should help you.

[pmerrill]

Tried the mod, not bad, from 46 to 55Mbits per sec with QoS, a 20% performance improvement.

I guess I can live without the stats and I don't really care about the reporting for a home network.

Many thanks!

[pbix]

There is no database of performance metrics for different routers so in reality no one can say of what you have observed is typical or not. But I will say that I am not surprised by what you have measured. It is true that Gargoyle requires more CPU than stock firmware that does not work.

Did you check to see if your CPU is maxing out? It sounds like it is but it could be something else so please check that an post the results.

If I were you and were blessed with a 100Mbps link I would try some different routers so I could take advantage of the many man years of software development that it took to create Gargoyle and the 100Mbps link.

I have a Buffalo WZR-HP-G300N and to me it seems way snappier than the 1043ND I had a few months back even though the specs of the two look similar. There are other high performance routers out there as well.

Then you could share what you learned and help others.

[mix]

It would be interesting if any particular process was using a large amount of the cpu during a bandwidth test. If you ssh into the router and run top, then run the bandwidth test from your computer, does top show the cpu usage spike for any particular process while the bandwidth test is running?

[pmerrill]

I looked at Top while doing a SpeedTest and the CPU numbers were nothing special, I think I rememebr seeing a 13% but in most cases it was about 4-5%. It appears that the CPU is doing very little work, mostly waiting for the next packet to come through.

Should point out, this was to a wired PC and not wireless.