Strange wifi performance (copying files to/from Windows 7 via SMB share)

I've seen some odd data transfer rates when copying large files between a Windows 7 PC and a Pi, over an SMB share. Instead of a fairly constant speed, sometimes/usually I got a rate which varied between about 0.5 and 80 Mbps as a sawtooth with a periodicity of about 20 seconds. For an Ethernet connection instead of wifi, the transfer rate was always constant.

formatting link
shows a fairly constant transfer; also shows the end of a previous transfer which was mostly a sawtooth, but which spontaneously started running at constant speed towards end of the transfer

formatting link
shows a transfer which was always a sawtooth

Any suggestions as to why it is happening? Is it a wifi "funny" (eg some sort of interference from a non-wifi source) or is the wifi adapter in the Pi intermittently behaving oddly? I realise that the Ethernet and wifi adapters in the Pi effectively go via the USB controller.

Background

----------------

I have a Window 7 PC and a Raspberry Pi 3B+.

The Win 7 was always connected to a TP Link 9980 router by Ethernet, at 1 Gbps speed (as shown by Task Manager | Networking). The router was not connected to ADSL/VDSL during the test - the phone cable was unplugged.

The Pi was connected either by Ethernet or Wifi, and the Wifi was either 2.4 GHz or 5 GHz.

There were no other wireless networks visible on Wifianalyzer app on an Android PC. I wasn't using a microwave oven (those sometimes put out a bit of 2.4 GHz shash).

A USB disk on the Pi was mounted as /home/pi/Videos and shared by Samba, and was accessible as a drive in Windows (net use r: \\win7\pi-share).

I measured file transfer read/write rates (a 1.3 GB file, with the transfer done as a copy/paste in Windows Explorer) between the Win 7 and the Pi using the various networking connections between the Pi and the router, always copying to/from the same SATA disc on the Windows PC.

The Pi and the router were about a foot apart, so signal strength was strong. And there were no other wifi networks.

Note that only *one* leg of the connection was wifi: the Windows 7 PC was always connected by Ethernet, to avoid wireless clashes between the two legs of the connection.

Results (for those who want the details!)

---------

Times (m:ss) for 1.32 GB file were:

Pi connected by Ethernet:

- 1:19 Win7 copies to Pi share (16.7 MB/sec, transfer rate about 160 Mbps, fairly constant))

- 1:17 Win7 copies from Pi share (17.1 MB/sec, transfer rate about 160 Mbps - fairly constant)

Pi connected by 2.4 GHz wifi

- Win 7 copies to Pi: 5:55 (sawtooth between 0.4 and 80 Mbps)

- Win 7 copies from Pi: 7:28 and 7:57 (two runs), transfer rate varies as sawtooth (about 20 second period) between 0.4 and 80 Mbps

Pi connected by 5 GHz wifi

- Win7 to Pi: 2:30 and 3:29 (first run was flat-out at about 90 Mbps, second run was sawtooth between 5 and 90 Mbps)

- Win7 from Pi: 2:30 and 4:27 (flat out at about 90 Mbps / sawtooth between

0.4 and 90 Mbps)

So the Pi is capable, at best, of copying at about 90 Mbps, at a fairly constant rate, over 5 GHz wifi. But usually the transfer runs much lower, with just occasional peaks to about 90 Mbps. Over 2.4 GHz, I did not see any flat-out transfers and the speed varied between 0.4 Mbps and 80 Mbps, for 5 GHz some transfers were flat out and others varied between about 5 and 90 Mbps.

Conclusion

--------------

Not surprisingly, best and most consistent rates are achieved with an Ethernet connection. 5 GHz gives higher speeds and sometimes flat-out (predictable) transfer; 2.4 GHz is a lot slower because it always spends a proportion of the time running at well below the peak speed.

The reason I was testing it was to work out whether I could manage with wifi rather than Ethernet cable or Ethernet-over-mains (powerline) from the Pi (in one room) to the router and Win 7 PC on the other side of the wall in the neighbouring room. The Pi needs to be in one room because it is used for recording TV and that's where the aerial and satellite cables are; the other room is my study. Wifi is unreliable and could take ages to copy recordings to my Win 7 "server", so it will have to be powerline - or else a hole drilled at skirting board level in the living room wall which will not get SWMBO approval ;-)

Reply to
NY
Loading thread data ...

The reported speed on both my Pi and my laptop varies between 1Mbps and

65Mbps. For no change in location of either.

Both use broadcomm chipsets.

disabling wifi and reenabling it on the laptop generally gets me back to

65Mbps.

I am remote from other dwellings in an RF quite place and screened from other RF by the metal in the house (GPS doesnt work inside the house, mobile phone is variable).

My general conclusion is that wifi with broacomm chipsets under linux sucks big time.

It might be instructive to try a USB wifi dongle.

--
Climate Change: Socialism wearing a lab coat.
Reply to
The Natural Philosopher

Personally powerline is evil, generates broad band radio frequency interference. Tell her that the Pi has limited storeage and that if she wants to watch recordings of the hole is required.

Or wait until she goes out for half a day and just do it. Only needs to be a 6 mm hole to get the cable through, check carefully for any other services hidden in the wall mind. If you have the abilty to fit RJ45 plugs do that to the free end. Slightly more acceptable might be a modular networking face plate on a surface box (or plasteboard flush box if the wall is platerboard) to just plug a patch lead into. Locate the hole box and cable route so it's out of sight and don't use a bright red cable... B-)

--
Cheers 
Dave.
Reply to
Dave Liquorice

Powerline works for one point to point. It sucks big time at adding more connections

--
Climate Change: Socialism wearing a lab coat.
Reply to
The Natural Philosopher

Is it worse than wifi in that respect? I've had fairly good multi-user performance on powerline, whereas wifi, even with devices right next to the router, is atrocious on my laptop, my pi and my desktop, especially if more than one is talking.

Am I right that two devices talking to each other via wifi and the router (both devices wif rather than Ethernet) will be significantly slower than one device by wifi and one by Ethernet, because both devices are competing for bandwidth?

What about wifi extenders (either dedicated or an old router in bridge mode). Even if there is only one device talking to the router (eg to the internet), is the performance worse (half?) if the wifi is connected to the extension rebroadcasted wifi than if it's connected directly to the router's own wifi?

For an extension network, is the best advice to give the extension a

*different* SSID to the router's wifi (eg "ROUTER_EXT") and to give it a *different*, non-overlapping channel? I've seen advice to put both networks on the *same* channel which I'd have thought would cause interference between the two.
Reply to
NY

When I tried two devices talking to my server simulatmneously speed dropped by a factor of about 10. This =was on TP link kit. I assumed their collision detection was shit.

Yup.

Dunno, Mi wifi extender is a router wired to ethernet...

Again. I dunno. I put mine on a different channel and different SSID

--
I would rather have questions that cannot be answered... 
...than to have answers that cannot be questioned 
 Click to see the full signature
Reply to
The Natural Philosopher

On Wed, 22 May 2019 10:33:55 +0100, "NY" declaimed the following:

"devices right next to the router" could be part of the problem. The received RF power levels could be causing front-end overload and distortion.

I'd presume the two (or more) are using different WiFi channels (there are only about 3 channels in the 2.4GHz WiFi band that can be used without overlapping bandwidth). That presumes your "extender" is hard-wired, and not a WiFi repeater.

Same channel probably expects clients to communicate with which ever node is the strongest -- without having to change channels. Note that using different SSID does mean the client has to disconnect from one SSID and negotiate a connection with the other (and if you are somewhere equidistant from both, might run the risk of sporadic signal loss/connect cycles).

--
	Wulfraed                 Dennis Lee Bieber         AF6VN 
	wlfraed@ix.netcom.com
Reply to
Dennis Lee Bieber

ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.