You are not logged in.

#226 2020-07-21 01:29:16

GaKu999
Member
From: US/Eastern
Registered: 2020-06-21
Posts: 696

Re: The IWD thread

Testing IWD+systemd-networkd...

Most of it’s standard behavior can be disabled according to the wiki and ‘man 5 iwd.config’.

@Trilby, it bite me as well, once IWD was stopped the interface was gone!

IWD alone with it’s builtin dhcpcd client works fine, maybe will use it when they implemented a global switch to NOT use the router DNS and NTP, I don’t want to tweak that value for every connection or do some ad-hoc hack to resolv.conf.

I guess I’m just lucky with my hardware, most different tools behave the same.

Found recently that (~1/4 rate) during the boot process IWD and networkd enter race condition (it’s called that?), IWD gets connected, but networkd complains about the interface being busy, doesn’t get a lease and just gives up, probably that’s where my luck ran out.

Not a big of a deal (~1/4 rate), but that’s scary on a WiFi dependent headless setup!

For now I’ll rather stay with wpa_supplicant+networkd but the CLI interface of IWD is good indeed! Will keep track of how it develops.

Last edited by GaKu999 (2020-07-22 01:02:20)


My reposSome snippets

Heisenberg might have been here.

Offline

#227 2020-07-21 20:55:47

mcloaked
Member
From: Yorkshire, UK
Registered: 2012-02-02
Posts: 1,222

Re: The IWD thread

I have been using iwd on two laptops for about a year without any issues until a few weeks ago.  Occasionally and seemingly at random I get a wireless disconnect which gives the following entries in the log on a search for wlan:

Jul 21 20:23:39 lenovo1 kernel: wlan0: deauthenticating from xx:xx:bf:xx:a6:xx by local choice (Reason: 3=DEAUTH_LEAVING)
Jul 21 20:23:39 lenovo1 systemd-networkd[443]: wlan0: Link DOWN
Jul 21 20:23:39 lenovo1 avahi-daemon[387]: Interface wlan0.IPv6 no longer relevant for mDNS.
Jul 21 20:23:39 lenovo1 avahi-daemon[387]: Leaving mDNS multicast group on interface wlan0.IPv6 with address fd00:abad:f00d:d00d::94.
Jul 21 20:23:39 lenovo1 avahi-daemon[387]: Interface wlan0.IPv4 no longer relevant for mDNS.
Jul 21 20:23:39 lenovo1 avahi-daemon[387]: Leaving mDNS multicast group on interface wlan0.IPv4 with address 10.0.0.94.
Jul 21 20:23:39 lenovo1 avahi-daemon[387]: Withdrawing address record for 2a00:23c6:1c83:2500:494f:9cf3:xxxx:4715 on wlan0.
Jul 21 20:23:39 lenovo1 avahi-daemon[387]: Withdrawing address record for fdaa:bbcc:ddee:0:494f:9cf3:3bbb:xxxx on wlan0.
Jul 21 20:23:39 lenovo1 avahi-daemon[387]: Withdrawing address record for 2a00:23c6:1c83:2500:e8b:fdff:febd:xxxx on wlan0.
Jul 21 20:23:39 lenovo1 avahi-daemon[387]: Withdrawing address record for fdaa:bbcc:ddee:0:e8b:fdff:febd:xxxx on wlan0.
Jul 21 20:23:39 lenovo1 avahi-daemon[387]: Withdrawing address record for fd00:abad:xxxx:xxxx::94 on wlan0.
Jul 21 20:23:39 lenovo1 avahi-daemon[387]: Withdrawing address record for 10.0.0.94 on wlan0.
Jul 21 20:23:39 lenovo1 systemd-networkd[443]: wlan0: Lost carrier
Jul 21 20:23:39 lenovo1 systemd-networkd[443]: wlan0: DHCP lease lost
Jul 21 20:23:40 lenovo1 systemd-networkd[443]: wlan0: Could not find device, waiting for device initialization: No such device
Jul 21 20:23:40 lenovo1 systemd-udevd[8915]: wlan0: Failed to get link config: No such device
Jul 21 20:23:40 lenovo1 systemd-networkd[443]: wlan1: Link UP

I am using iwd, systemd-networkd, and the interface is named (normally) wlan0  - the journal shows that the kernel is deauthenticating from the AP, and then systemd-networkd decides the link is down. udevd then decides there is no such device - presumably wlan0 - and then systemd-networkd makes wlan1 become up.  It then seems to connect and get an address.  All systemd units are then showing none failed, but I have no network running.

My /etc/systemd/network/25-wireless.network file contains (with adjustments of addresses for security):

[Match]
#Name=wlp4s0
MACAddress=xx:xx:xx:xx:xx:xx

[Network]
Description=lenovo1w

#DHCP=ipv4
DHCP=yes
IPv6PrivacyExtensions=true


[Address]
# ULA
Address=fd00:xxxx:xxxd:xxxd::94/64
#

[DHCP]
RouteMetric=20

So connects to the matched mac address which usually works fine. 

However if I try to restart the associated unit services iwd, or systemd-networkd the network still does not come up and I can't ping sites as when it was originally connected to wlan0.  I then have to resort to rebooting, when everything then returns to normal and wireless works fine again.

Has anyone else come across this issue and if so is there a workaround?  Thanks.

Edit: I decided to try to use a .link file to set the interface name:

$ cat /etc/systemd/network/10-wlan0.link 
[Match]
MACAddress=xx:xx:xx:xx:xx:xx

[Link]
Description=Wireless interface
Name=wlan0

and also then use the name insread of the MACAdress match in the associated network file:

$ cat /etc/systemd/network/25-wireless.network
[Match]
Name=wlan0

[Network]
Description=lenovo1w

#DHCP=ipv4
DHCP=yes
IPv6PrivacyExtensions=true


[Address]
# ULA
Address=fd00:abad:f00d:d00d::94/64
#

[DHCP]
RouteMetric=20

On reboot the wireless comes up fine so I will see if this shows the same periodic failures as happened until this latest change.

It does puzzle me that in the unit status at:

$ sudo systemctl status systemd-networkd
● systemd-networkd.service - Network Service
     Loaded: loaded (/usr/lib/systemd/system/systemd-networkd.service; enabled; vendor preset: enabled)
     Active: active (running) since Tue 2020-07-21 22:13:28 BST; 11min ago
TriggeredBy: ● systemd-networkd.socket
       Docs: man:systemd-networkd.service(8)
   Main PID: 441 (systemd-network)
     Status: "Processing requests..."
      Tasks: 1 (limit: 9419)
     Memory: 3.1M
     CGroup: /system.slice/systemd-networkd.service
             └─441 /usr/lib/systemd/systemd-networkd

Jul 21 22:13:28 lenovo1 systemd-networkd[441]: wlan0: Could not find device, waiting for device initialization: N>
Jul 21 22:13:28 lenovo1 systemd-networkd[441]: wlan0: Link DOWN
Jul 21 22:13:28 lenovo1 systemd-networkd[441]: wlan0: Link UP
Jul 21 22:13:28 lenovo1 systemd-networkd[441]: enp3s0: IPv6 successfully enabled
Jul 21 22:13:28 lenovo1 systemd-networkd[441]: enp3s0: Link UP
Jul 21 22:13:28 lenovo1 systemd-networkd[441]: wlan0: Connected WiFi access point:

systemd-networkd still needs to wait for device initialisation - so perhaps I need to tell it to wait for the device to be up in its service file - sp perhaps it needs to wait for udevd to be up?

Last edited by mcloaked (2020-07-21 21:27:21)


Mike C

Offline

#228 2020-07-22 01:23:44

GaKu999
Member
From: US/Eastern
Registered: 2020-06-21
Posts: 696

Re: The IWD thread

@mcloaked, idk how to fix that, but if the issue is related to IWD eating the interface you may try:

Create/edit the file /etc/iwd/main.conf and add:

[General]
UseDefaultInterface=true

Don’t duplicate [General] if you have one already.

If you also wish to have the default names for the wlan interfaces (instead of wlanx) use:
‘ln -sf /dev/null /etc/systemd/network/80-iwd.link’

For more info you may look at ‘man 5 iwd.config’ and the wiki

Also to note, I once had similar issues with a defective router...so you may want take a look at that...or don’t if you are sure that’s not the case.


My reposSome snippets

Heisenberg might have been here.

Offline

#229 2020-07-22 17:09:43

mcloaked
Member
From: Yorkshire, UK
Registered: 2012-02-02
Posts: 1,222

Re: The IWD thread

Thanks - I am aware of going to legacy but would prefer not to do that.  I will soak test the laptop with the additional link file that matches the mac address to the wlan0 name and see if the problem recurs or not - hopefully if the interface goes down  I will see if it still gets the same name when it reconnects or not. I will post again in a few days once I have thoroughly tested that. The router is not an issue - and no other devices (phones etc) have a deauth issue.


Mike C

Offline

#230 2020-07-22 18:08:43

GaKu999
Member
From: US/Eastern
Registered: 2020-06-21
Posts: 696

Re: The IWD thread

mcloaked wrote:

Thanks - I am aware of going to legacy but would prefer not to do that.

Idk what do you mean by legacy but if it’s the interface name that’s the second option, the first one of ‘/etc/iwd/main.conf’ makes IWD NOT eat the interface like he’s the owner of the place, meaning that it will not disappear when it stops or whatever. Since your issue is related to wlan0 vanishing into the void, that MIGHT at least solve part of the issue if not all of it.


My reposSome snippets

Heisenberg might have been here.

Offline

#231 2020-07-22 19:00:49

mcloaked
Member
From: Yorkshire, UK
Registered: 2012-02-02
Posts: 1,222

Re: The IWD thread

GaKu999 wrote:
mcloaked wrote:

Thanks - I am aware of going to legacy but would prefer not to do that.

Idk what do you mean by legacy but if it’s the interface name that’s the second option, the first one of ‘/etc/iwd/main.conf’ makes IWD NOT eat the interface like he’s the owner of the place, meaning that it will not disappear when it stops or whatever. Since your issue is related to wlan0 vanishing into the void, that MIGHT at least solve part of the issue if not all of it.

Legacy is the description here https://wiki.archlinux.org/index.php/Iw … _available

Presumably if you do make that change to the iwd main.conf then the interface (in my case) would get the name wlp1s0 or similar but not wlan0?


Mike C

Offline

#232 2020-07-22 19:07:58

GaKu999
Member
From: US/Eastern
Registered: 2020-06-21
Posts: 696

Re: The IWD thread

mcloaked wrote:
GaKu999 wrote:
mcloaked wrote:

Thanks - I am aware of going to legacy but would prefer not to do that.

Idk what do you mean by legacy but if it’s the interface name that’s the second option, the first one of ‘/etc/iwd/main.conf’ makes IWD NOT eat the interface like he’s the owner of the place, meaning that it will not disappear when it stops or whatever. Since your issue is related to wlan0 vanishing into the void, that MIGHT at least solve part of the issue if not all of it.

Legacy is the description here https://wiki.archlinux.org/index.php/Iw … _available

Presumably if you do make that change to the iwd main.conf then the interface (in my case) would get the name wlp1s0 or similar but not wlan0?

My bad, didn't though of IWD legacy mode...

If you make that change but don't touch the '80-iwd.link' file or mask it in '/etc/systemd/network/80-iwd.link' to '/dev/null' as the second option shows, then you will still see wlan0.
if you do BOTH then you will have wlp1s0.
if you only mask '80-iwd.link' it will also stay as wlan0 because IWD renames it, ignoring '80-iwd.link'.


My reposSome snippets

Heisenberg might have been here.

Offline

#233 2020-07-22 19:12:53

mcloaked
Member
From: Yorkshire, UK
Registered: 2012-02-02
Posts: 1,222

Re: The IWD thread

OK thanks.  I will see how the use of the link file that sets the normal name wlan0 as a match to the interface mac address works for stability. So far today there have been no de-auth events.  If further testing shows it still gives issues then I will try the masking of the file you suggest.


Mike C

Offline

#234 2020-07-22 19:17:50

GaKu999
Member
From: US/Eastern
Registered: 2020-06-21
Posts: 696

Re: The IWD thread

mcloaked wrote:

OK thanks.  I will see how the use of the link file that sets the normal name wlan0 as a match to the interface mac address works for stability. So far today there have been no de-auth events.  If further testing shows it still gives issues then I will try the masking of the file you suggest.

Note that I said it if you prefer to use wlp1s0 as a name, on my case I happily use wlp5s0 with wpa_supplicant, because it's stable and doesn't change in unexpected ways, like starting with an USB WiFi dongle and somehow my in board device being wlan1!

That's totally optional and biased on preferences and/or use cases.

Last edited by GaKu999 (2020-07-22 19:25:02)


My reposSome snippets

Heisenberg might have been here.

Offline

#235 2020-07-22 20:47:36

gothmog123
Member
Registered: 2012-10-31
Posts: 120

Re: The IWD thread

Trilby wrote:

Oh sorry, I missread your initial report that IWD+Networkmanager was slower than IWD alone.  That would suggest my interpretation may be incorrect.  We need more data points to make reasonable inferences - I can only say what happened on my system.

Have you tried IWD with dhcpcd or dhclient?

Interestingly dhcpcd gives me good, iwd-like speeds. But network managers don't support dhcpcd and I want to use a network manager so I can control wifi from a tray icon.

Offline

#236 2020-07-23 09:41:50

rsmarples
Member
Registered: 2009-05-12
Posts: 287

Re: The IWD thread

gothmog123 wrote:

Interestingly dhcpcd gives me good, iwd-like speeds. But network managers don't support dhcpcd and I want to use a network manager so I can control wifi from a tray icon.

dhcpcd is a network manager for everything other than link setup.
There is the dhcpcd-ui project which has tray icons for GTK+ and Qt - I believe this is packaged in AUR.

Last edited by rsmarples (2020-07-23 09:42:11)

Offline

#237 2020-07-23 09:44:19

rsmarples
Member
Registered: 2009-05-12
Posts: 287

Re: The IWD thread

Oh, and just to clarify something here - dhcpcd won't affect the speed of any connections and it's extermely unlikely other DHCP clients would either.
This is all link management where the bulk of the work is in the wireless chipset, then the driver and what it can use (hw crypto, etc) and then the link management layer (wpa_supplicant or iwd).

Offline

#238 2020-07-23 14:31:22

GaKu999
Member
From: US/Eastern
Registered: 2020-06-21
Posts: 696

Re: The IWD thread

rsmarples wrote:

Oh, and just to clarify something here - dhcpcd won't affect the speed of any connections and it's extermely unlikely other DHCP clients would either.
This is all link management where the bulk of the work is in the wireless chipset, then the driver and what it can use (hw crypto, etc) and then the link management layer (wpa_supplicant or iwd).

That was indeed needed, most of the speed differences are because IWD and wpa_supplicant have differences on the time of module loading.
In the case of @Trilby IWD loaded needed modules for his hardware that wpa_supplicant missed.
That can also happen the other way around, wpa_supplicant loading modules that IWD misses.
But the same remains, it's a hardware dependent issue.

This may be inaccurate, because I lack the knowledge of HOW this process works at the lower level and what is failing to detect what modules to load in both cases, or even if it's external to them...


My reposSome snippets

Heisenberg might have been here.

Offline

#239 2020-07-23 14:38:13

Trilby
Inspector Parrot
Registered: 2011-11-29
Posts: 29,442
Website

Re: The IWD thread

Please keep in mind my data are purely anecdotal based on results one one machine.  Rsmarples would know the inner workings better - and it is logical that the dhcp client would not / could not affect connection speed.  Though this really does beg the question of how networkmanager manages to screw it up on your system as it *should* leave all the lower level link management to iwd.  Can we assume you've replicated this effect a few times on fresh boot ups?

In addiiton to he dhcpcd gui options noted above, there connman and netctl each have GUI front-ends available some of which include tray icons.  If a gui w/ tray icon is the only thing holding you to NM, there are alternatives you could try.


"UNIX is simple and coherent..." - Dennis Ritchie, "GNU's Not UNIX" -  Richard Stallman

Offline

#240 2020-07-23 14:44:47

gothmog123
Member
Registered: 2012-10-31
Posts: 120

Re: The IWD thread

rsmarples wrote:

Oh, and just to clarify something here - dhcpcd won't affect the speed of any connections and it's extermely unlikely other DHCP clients would either.
This is all link management where the bulk of the work is in the wireless chipset, then the driver and what it can use (hw crypto, etc) and then the link management layer (wpa_supplicant or iwd).

sorry but for me the sitation is absolutely clear: iwd with its own dhcp client gives better speeds than using anything else. also if I launch iwd with its own client then switch to something else, the speed doesn't decrease until I reboot. there is some kind of configuration magic that iwd applies, I'm absolutely certain

Last edited by gothmog123 (2020-07-23 14:45:08)

Offline

#241 2020-07-23 14:49:23

gothmog123
Member
Registered: 2012-10-31
Posts: 120

Re: The IWD thread

Trilby wrote:

Please keep in mind my data are purely anecdotal based on results one one machine.  Rsmarples would know the inner workings better - and it is logical that the dhcp client would not / could not affect connection speed.  Though this really does beg the question of how networkmanager manages to screw it up on your system as it *should* leave all the lower level link management to iwd.  Can we assume you've replicated this effect a few times on fresh boot ups?

In addiiton to he dhcpcd gui options noted above, there connman and netctl each have GUI front-ends available some of which include tray icons.  If a gui w/ tray icon is the only thing holding you to NM, there are alternatives you could try.

I think dhcpcd gui can't do IWD, no? I'm specifically looking to use IWD.

Last edited by gothmog123 (2020-07-23 14:49:46)

Offline

#242 2020-07-23 14:50:46

GaKu999
Member
From: US/Eastern
Registered: 2020-06-21
Posts: 696

Re: The IWD thread

gothmog123 wrote:
rsmarples wrote:

Oh, and just to clarify something here - dhcpcd won't affect the speed of any connections and it's extermely unlikely other DHCP clients would either.
This is all link management where the bulk of the work is in the wireless chipset, then the driver and what it can use (hw crypto, etc) and then the link management layer (wpa_supplicant or iwd).

sorry but for me the sitation is absolutely clear: iwd with its own dhcp client gives better speeds than using anything else. also if I launch iwd with its own client then switch to something else, the speed doesn't decrease until I reboot. there is some kind of configuration magic that iwd applies, I'm absolutely certain

Attempt a diff with lsmod to see what different modules IWD decides to load, also the are other configuration values that need to be adjusted sometimes depending on the hardware...
There's a lot of information in the wiki, covering many use cases.


My reposSome snippets

Heisenberg might have been here.

Offline

#243 2020-07-23 15:02:22

Trilby
Inspector Parrot
Registered: 2011-11-29
Posts: 29,442
Website

Re: The IWD thread

gothmog123 wrote:

sorry but for me the sitation is absolutely clear: iwd with its own dhcp client gives better speeds than using anything else.

No.  That is not true.  It is most certainly not absolutely clearly true.  Several reports above note that IWD + dhcpcd or IWD + dhclient acheive the better speeds too.  In fact, I acheived the better speeds even without IWD simply by loading the kernel modules it would have loaded.  Rsmarples claim was that the dhcp client should not be relevant, only the link is relevant for the speed.  All data in this thread so far supports that *except* for one reported case of IWD+NM not acheiving the speed improvement.

Last edited by Trilby (2020-07-23 15:03:30)


"UNIX is simple and coherent..." - Dennis Ritchie, "GNU's Not UNIX" -  Richard Stallman

Offline

#244 2020-07-23 18:09:39

gothmog123
Member
Registered: 2012-10-31
Posts: 120

Re: The IWD thread

Trilby wrote:
gothmog123 wrote:

sorry but for me the sitation is absolutely clear: iwd with its own dhcp client gives better speeds than using anything else.

No.  That is not true.  It is most certainly not absolutely clearly true.  Several reports above note that IWD + dhcpcd or IWD + dhclient acheive the better speeds too.  In fact, I acheived the better speeds even without IWD simply by loading the kernel modules it would have loaded.  Rsmarples claim was that the dhcp client should not be relevant, only the link is relevant for the speed.  All data in this thread so far supports that *except* for one reported case of IWD+NM not acheiving the speed improvement.


for me connman is also affected, along with NM. it's true that dhcpcd is not affected

Offline

#245 2020-07-23 21:00:14

rsmarples
Member
Registered: 2009-05-12
Posts: 287

Re: The IWD thread

gothmog123 wrote:
Trilby wrote:

Please keep in mind my data are purely anecdotal based on results one one machine.  Rsmarples would know the inner workings better - and it is logical that the dhcp client would not / could not affect connection speed.  Though this really does beg the question of how networkmanager manages to screw it up on your system as it *should* leave all the lower level link management to iwd.  Can we assume you've replicated this effect a few times on fresh boot ups?

In addiiton to he dhcpcd gui options noted above, there connman and netctl each have GUI front-ends available some of which include tray icons.  If a gui w/ tray icon is the only thing holding you to NM, there are alternatives you could try.

I think dhcpcd gui can't do IWD, no? I'm specifically looking to use IWD.

Ah yes, this is true.
dhcpcd-ui only handles wpa_supplicant for setting up wpa_supplicant.conf.

I don't have too much motivation to write support for IWD because the majority of my work occurs on various BSD families.
I'm mainly here to support arch dhcpcd users where I can smile

Offline

#246 2020-07-23 21:16:38

mcloaked
Member
From: Yorkshire, UK
Registered: 2012-02-02
Posts: 1,222

Re: The IWD thread

I had a recurrence of the issue I reported earlier in this thread - and decided to put in an arch bug report at https://bugs.archlinux.org/task/67373 which was closed as not a bug but a config issue! Anyway I have now found an upstream bug on the systemd github site which looks rather related at  https://github.com/systemd/systemd/issues/15691 - so I am not sure that it is entirely merely a config issue!

Last edited by mcloaked (2020-07-23 21:16:53)


Mike C

Offline

#247 2020-07-24 00:39:33

GaKu999
Member
From: US/Eastern
Registered: 2020-06-21
Posts: 696

Re: The IWD thread

mcloaked wrote:

I had a recurrence of the issue I reported earlier in this thread - and decided to put in an arch bug report at https://bugs.archlinux.org/task/67373 which was closed as not a bug but a config issue! Anyway I have now found an upstream bug on the systemd github site which looks rather related at  https://github.com/systemd/systemd/issues/15691 - so I am not sure that it is entirely merely a config issue!

Don't evolve into "bikeshed" please, also in the bug report I see you did not touch '/etc/iwd/main.conf'...
Why ask for support if you are just going to ignore it?

Instead, try a more "vanilla" wireless setup with:

[Match]
Name=wlp*
Name=wlan*

[Network]
DHCP=yes
IPv6PrivacyExtensions=yes

[DHCP]
RouteMetric=1024

And the change I mentioned to '/etc/iwd/main.conf' to see if the issue replicates...

Last edited by GaKu999 (2020-07-24 00:47:03)


My reposSome snippets

Heisenberg might have been here.

Offline

#248 2020-07-24 17:10:20

mcloaked
Member
From: Yorkshire, UK
Registered: 2012-02-02
Posts: 1,222

Re: The IWD thread

GaKu999 wrote:
mcloaked wrote:

I had a recurrence of the issue I reported earlier in this thread - and decided to put in an arch bug report at https://bugs.archlinux.org/task/67373 which was closed as not a bug but a config issue! Anyway I have now found an upstream bug on the systemd github site which looks rather related at  https://github.com/systemd/systemd/issues/15691 - so I am not sure that it is entirely merely a config issue!

Don't evolve into "bikeshed" please, also in the bug report I see you did not touch '/etc/iwd/main.conf'...
Why ask for support if you are just going to ignore it?

Instead, try a more "vanilla" wireless setup with:

[Match]
Name=wlp*
Name=wlan*

[Network]
DHCP=yes
IPv6PrivacyExtensions=yes

[DHCP]
RouteMetric=1024

And the change I mentioned to '/etc/iwd/main.conf' to see if the issue replicates...

Ok GaKu999

I am running now with both the wlan* and wlp* matches in the network file as per your suggestion, as well as the two line /etc/iwd/main.conf containing UseDefaultInterface=true, plus I took out the .link file. I will now run an extended time to test the wireless connection. Running only a short time there is nothing in the logs that looks bad so far and I will see if I get any dropped connections after an hour or two.


Mike C

Offline

#249 2020-08-01 14:18:03

Trilby
Inspector Parrot
Registered: 2011-11-29
Posts: 29,442
Website

Re: The IWD thread

I have to "publish" a retraction of my previous notes about speed increases with iwd and it being related to specific kernel modules.

Since then I've been exploring this more and more and been getting inconsistent results.  Throughout all of it I had been using `speedtest-cli` from [community].  I took for granted that the numbers obtained from that tool were meaningful.  It seems they are not.  I started seeing wildly different results from back-to-back runs of speedtest-cli.  Previously I assumed fluctuations were real, i.e., my wireless connection really did speed and slow, and so I intended to take many readings with speedtest-cli to get a better average estimate.

But this is where I realized there was far far (far far far) more variability of speeds within condition (iwd vs wpa_supplicant) than between them.  This is when I became suspicious of speedtest-cli and tried something different.  I have a very lightly used VPS server on the other side of the country: I wrote a curl script to download a large file from it and report the observed download speed.  The results surprised me: my network connection bandwidth is incredibly stable and consistent.  It would seem the wild variability in results I received from speedtest-cli were from bottlenecks either in the servers it connected to or en route to them.  With a consistent target server (and likely similar or identical route) every replication gave consistent readings for my connection speed.

I replicated this with repeated curl downloads of a couple large packages from the kernel.org mirrors, and the speed of these replicates was also pretty consistent (but different from my tests with my VPS as the target).

It would, therefore seem that I was right to be surprised that my local wireless connection could be a bandwidth limiting step.  Not only is my wlan not bandwidth limiting, my local ISP is also not bandwidth limiting.  None of these tests are providing any metric of my connection speed, but rather a metric of the target at the other end (especially in the case of speedtest-cli where this target varies each time).

I also confirmed with my test downloading from my VPS did not differ at all between iwd and wpa_supplicant.

So the speed differences I previously reported were a fluke.  I also noted subjective but notably faster loading of videos - but that didn't remain.  I was rebooting to test IWD initially, and it seems on a fresh reboot sometimes videos do load a bit faster regardless of the networking tool in use.

Moral of the story: 1) I see zero evidence that iwd provides faster downloads than wpa_supplicant and 2) speedtest-cli is a fucking joke.

IWD does seem to establish connections faster, but I was also noting frequent disconnect-reconnect cycles with it that I've never seen with wpa_supplicant.  It reconnected so fast that it was never a burden, but it was odd.

Last edited by Trilby (2020-08-01 17:48:31)


"UNIX is simple and coherent..." - Dennis Ritchie, "GNU's Not UNIX" -  Richard Stallman

Offline

#250 2020-08-01 17:44:50

ondoho
Member
Registered: 2013-04-30
Posts: 692
Website

Re: The IWD thread

^ Pity, I was hoping to benefit. But thanks anyhow. I might still test this if I figure out how...

Offline

Board footer

Powered by FluxBB