[Beta] Asuswrt-Merlin 380.59 Beta 2 is available

[RMerlin]

i think those test from speedtest or dslreport sites is not real measuring. i tested now ten times just with disabled QoS and got very different speedtest, bufferbloat and quality results....we need one better way to real test our network quality.....

Those online speedtest are definitely broad, generic tests that might not give the whole story, but can be used as a basis to give you a hint at what is going on.

What would be a better test would be for someone to run his own multiple simultaneous streams, and then monitor latency both by running a parallel ping, or by checking how long the data takes to get from point A to B (latency-wise, not transfert-time). That's probably what DSLReports does, except their server might also have varying performance.

Real-life tests would be good as well. Start a VoIP communication, then start multiple parallel downloads and uploads and see how long until your VoIP performance degrades.

 

[RMerlin]

Was the AC66 binary updated to 2695 or is that still not available? They have the beta firmware up on https://www.asus.com/support/Download/11/2/0/84/LzShv8ma7TrQB4eO/30/

The RT-AC66U has been supported since beta 1.

 

[RMerlin]

Anybody else seeing this behaviour on 2.4 GHz? the 5GHz seems a lot more stable

Try doing a factory default reset and manually reconfiguring your router.

 

[westmc]

@RMerlin I'm experiencing an issue with Samba on my RT-AC68P with 380.59 beta 2. I had the same issue when I was briefly on 380.58. I did not have the issue with 380.57 or any of the stock versions including 380.2695, so I'm assuming it is something specific to the Merlin branch. I'm hoping you can help me diagnose the problem.

Here is the setup. WD Elements 4TB HDD attached through USB 3.0. One partition is ext4 used for backups over Samba and the other partition is HFS+ for backups over TimeMachine. I only have metrics for my Windows backups using Acronis so I don't know if the issue also affects TimeMachine. I have daily backups run. One day a week, it is a full backup and the rest of the days it is a differential backup. Under 380.58 and 380.59 the full backup went from 1.5 hours to 10 and the differentials went from 3 minutes to 30. I was hoping the issue was just with 380.58 because I also experienced other issues but I am now seeing it with 380.59 beta 2 after going back to 380.2695 and seeing the problem go away. What kind of logs or info can I get you to help figure out what the issue is?

For my router config, it is super simple. I reset to factory every time and I only change a few things. In terms of disk related changes, I disable that USB 3.0 interference setting, I turn on TimeMachine, I enable Samba with a non-default device name, simple share naming and force as master browser, and I enable the media server with it set to one specific folder.

Nothing has changed with my config in months and it is 100% reproducible and there are no issues that I have seen with the speed of my connection between firmware versions so it must be something internal to the router's samba. As I type this, a full backup has been going on for over 6 hours. Windows task manager shows an oddly steady 32.1 Mbps coming from Acronis and the router GUI shows 1 core at 6% and the other at 40%. Memory is at 89%. Does the 32.1 Mbps stand out to you? It is odd that it is so steady.

Here is another metric. I have used CrystalDiskMark in the past to measure Samba performance and it has typically been about 30MBs R/W over wireless and 55MBs R/W over wired. Under 380.59 beta 2 I am getting 21MBs R and 3 MBs W under both wireless and wired.

 

[___Lele___]

Try doing a factory default reset and manually reconfiguring your router.

I will. I also narrowed it down to some interference with USB 3.0 which raises the noise floor a lot. in some cases even to -70dBm which is a lot.

Did you try to play with the "Reducing USB 3.0 interference" option in the firmware? Mine is enabled by default buy I wonder if there are some side effects which would explain the dropouts.

Do you know of any low-level logs for the radios? I remember on Cisco routers I could do that through the CLI. I'd like to have a clearer picture of what is going on from PHY level

 

[LouisvilleUK]

The OpenVPN fix in 380.59 beta 2 fixed the issue I was having in 380.58 when using policy-based mode with “exclusive” DNS config. In 380.58 had to revert back to using DNSFilter and “disable” DNS config. With 380.59 beta 2 I’m able to use as advertised!

Great work and thank you!

 

[RMerlin]

Did you try to play with the "Reducing USB 3.0 interference" option in the firmware? Mine is enabled by default buy I wonder if there are some side effects which would explain the dropouts.

Enabling this option will downgrade the USB port to USB 2.0, which eliminates the radio interference. It's definitely not ideal.

Do you know of any low-level logs for the radios? I remember on Cisco routers I could do that through the CLI. I'd like to have a clearer picture of what is going on from PHY level

Perhaps, but not having access to Broadcom's SDK, I have no idea if or how.

 

[RMerlin]

Here is another metric. I have used CrystalDiskMark in the past to measure Samba performance and it has typically been about 30MBs R/W over wireless and 55MBs R/W over wired. Under 380.59 beta 2 I am getting 21MBs R and 3 MBs W under both wireless and wired.

Some people saw an increase with the Samba changes in 380.59, and others saw a decrease. I'll probably just revert the change altogether, so performance will be in line with the stock firmware's (minus the performance loss of Samba 3.6 vs Samba 3.0.25b).

However this wouldn't explain any issue from 380.58, where there has been no change to Samba.

 

[barbara]

You will have to diff the whole target.mak profile to figure out which setting you are missing, or you have set differently from Asus. Settings such as DEBUGFS will cause kernel symbol changes if they are changed, so they must match what's used by Asus.

My custom mods were working fine with version 380.58. Then Asus introduced a change in version 380.59 by adding two new prebuilt "o" files: ate.o and dualwan.o, which depend on BWDPI, I think.

So, I believe the problem is caused by my removal of BWDPI from the firmware and that I may have to compile it back in if I want the Tuxera filesystem drivers to work.

The two new "o" files depend on BWDPI, for some reason that I can't explain. However, since the source code for them is given in ate.c and dualwan.c, I did remove the prebuilt ate.o and dualwan.o. Everything compiles fine, except now I get those missing symbol errors in the Tuxera filesystem drivers.

I will try compiling BWDPI back in and see what happens.

 

[westmc]

Some people saw an increase with the Samba changes in 380.59, and others saw a decrease. I'll probably just revert the change altogether, so performance will be in line with the stock firmware's (minus the performance loss of Samba 3.6 vs Samba 3.0.25b).

However this wouldn't explain any issue from 380.58, where there has been no change to Samba.

I definitely had the same issue with 380.58. Can you point me to how to revert the samba changes and I can try it out to confirm whether that is the cause of the problem?

 

[westmc]

I definitely had the same issue with 380.58. Can you point me to how to revert the samba changes and I can try it out to confirm whether that is the cause of the problem?

I can also try compiling myself if you just give a patch or tell me what to change.

 

[vibroverbus]

Ran a series of dslreports with different configs and found minimal result differences between them.
Testing was done when network was not being loaded, so not real ideal test of QoS.
Bottom line not convinced any of below is statistically significant, nor am confident in dslreports testing effectiveness, nor in it being valid on less than highly loaded network.
But. For what its worth, here it is.

Am going to run the last config listed for a while and see how it goes.
I do tend to seem to have latency creep up over time and really slow responses when I know there's video streaming and other things going on, will be good to see what the results are with different loading.

Router: AC-3200
Figures below average of 3 or more runs

QoS settings tested:

157/11 bandwidth (connection is generally 180/12)
A few QoS rules I had in place prioritizing by client MAC address - but near zero network load while testing so doesn't reflect a real QoS active prioritization
Note all results run happily higher than the entered bandwidth which is odd - in previous routers with DD-WRT and QoS on, I would easily and reliably see the 'hard cap' of the QoS setting.
The Asus implementation seems to let traffic exceed the settings as long as there is no bottleneck?​

Base case QoS off - default txqueue = 1000
180/12 - provider hard caps the up much more consistently than the down which seems to be able to burst, has always been this way - never see an upload test that isn't 12M
70-100ms for dslreports top list
dslreport scores B/A/A

Q0S off - txqueue = 100
178/12 perhaps tiny decline in download speed but not substantial
50-70 latencies - marginally better
Dslreports B/A+/A (up on 'quality')

Q0S on - sfq w/ default txqueue = 1000
178/12 roughly same as above
50-60 latencies - marginally better
Dslreports B/A+/A (same as above)

Q0S on - fq_codel w/ default txqueue = 1000
176/12 little slower still
45-60 latencies - marginally better
Dslreports B/A+/A (same as above)

Q0S on - fq_codel w/ txqueue = 8/40/40
179/12 faster than default txq
60-80 latencies - back down again
Dslreports B/A+/A (same as above)

Q0S on - fq_codel w/ txqueue = 100
180/11 faster than default txq
75-100 latencies - slower again
Dslreports B/A+/A (same as above)

 

[dfran1]

RT-AC5300
I am having problems with my 2.4GHz band.
It seems like it is in and out.
I was on the alpha 1 that was perfect, I factory defaulted and used nvram save r23 to restore my settings.
I will try the 20 / 40 settings as ___Lele___ said fixed it.
I do have smart connect as I did before, but the nvram r23 did not copy that settings and maybe is not 100% ready for the 5300.
As for the QOS, I use Adaptive on my 5300.

 

[Theliel]

Hi RMerlin

My reports with QoS, RT-AC56u Fiber 30/3, 95%/90% cap.

Adaptive give me the best overall result
Trad. SFQ/Codel similar results, with good upload bufferbloat (better than Adaptive) but bad on download side
Trad. fq_codel give me the best results on upload test (bufferbloat), but again fail on Download, much worse than Adaptive

 

[LBJPilot]

I installed beta 2 yesterday. I am having problems with the traffic analyzer -> Statistics.. Was not showing any apps or clients... I erased the /jffs/traffic.db file. (had turned off the flag first)... turned in back on, a new file was created with 0 bytes, on the top of the hour the cpu was busy for short time, and then the stats showed up... Today I checked and again the stats is showing nothing. See picture below... I did check the file via ssh and it was recently updated.

Any suggestions? Are others having issues.

 

[RMerlin]

My custom mods were working fine with version 380.58. Then Asus introduced a change in version 380.59 by adding two new prebuilt "o" files: ate.o and dualwan.o, which depend on BWDPI, I think.

So, I believe the problem is caused by my removal of BWDPI from the firmware and that I may have to compile it back in if I want the Tuxera filesystem drivers to work.

The two new "o" files depend on BWDPI, for some reason that I can't explain. However, since the source code for them is given in ate.c and dualwan.c, I did remove the prebuilt ate.o and dualwan.o. Everything compiles fine, except now I get those missing symbol errors in the Tuxera filesystem drivers.

I will try compiling BWDPI back in and see what happens.

One of your symbol error is probably because you are missing BCM_RECVFILE=y in your target profile.

Also look for DEBUGFS - that's one flag tbat can also create symbol issues. You will have to diff your config against the stock firmware's own. Anything that touches the kernel can cause issues.

 

[RMerlin]

I definitely had the same issue with 380.58. Can you point me to how to revert the samba changes and I can try it out to confirm whether that is the cause of the problem?

You will have to remove this line from Samba:

socket options = IPTOS_LOWDELAY TCP_NODELAY SO_RCVBUF=65536 SO_SNDBUF=65536

I can't think of anything that would have changed with 380.58 however. That change was only with 380.59.

 

[RMerlin]

I can also try compiling myself if you just give a patch or tell me what to change.

The change is on Github, tho you can mimic it with a postconf script.

 

[RMerlin]

I installed beta 2 yesterday. I am having problems with the traffic analyzer -> Statistics.. Was not showing any apps or clients... I erased the /jffs/traffic.db file. (had turned off the flag first)... turned in back on, a new file was created with 0 bytes, on the top of the hour the cpu was busy for short time, and then the stats showed up... Today I checked and again the stats is showing nothing. See picture below... I did check the file via ssh and it was recently updated.

Any suggestions? Are others having issues.

Open your web browser console and look for any Javascript error.

Also give it 20-40 secs if you have a lot of traffic.

 

[LBJPilot]

Well both safari as well as chrome both do not display the data. On safari i get an invalid css property, on chrome I get a different error. Although I have seen the one on safari before and it does not cause the issue since I have seen it work OK even though I get that error. Never looked before on chrome. I think there is something else going on...

Copied both pics to post