Hello, I am reporting this bug, though it is well known for years. In general I've experienced poor wifi performance with all Intel cards, starting from 3945abg, 4965agn and now Intel 7260 Dual-Band AC.
The old cards were on a ThinkPad T61 system, 7260 is on a brand new ThinkPad Edge E540.
So here's the problem:
Wireless conection is slow, there's a unusual ping while browsing web pages (tried with all browsers - no luck), and sometimes drop-outs in connection.
There's a partial fix by disabling Wireless "N" mode, but this is irrational, because it limits the network to only 54mbits speed.
Details about the system:
Hardware:
ThinkPad E540
CPU: Intel core i5-4200m (Haswell)
Wireless adapter: Intel 7260 Dual-Band AC
Software:
PCLinuxOS 2014.07
Kernel: 3.15.5 (though it was the same with old kernels)
Kernel module: iwlwifi
On Tue, Jul 29, 2014 at 09:56:46AM -0500, Nate Carlson wrote:
> On Tue, 29 Jul 2014, Larry Finger wrote:
> >You should try the experimental firmware file from
> >https://git.kernel.org/cgit/linux/kernel/git/egrumbach/linux-firmware.git/plain/iwlwifi-7260-9.ucode?h=Core6.
> >
> >I am currently testing it and getting throughput of 60 Mbps on the 2.4 GHz band,
> >and 90 Mbps at 5 GHz.
I'm getting 1...5 Mbs with 7260 AC (Thinkpad Yoga) in 2.4 GHz band
in the office (11n AP), more decent speed at home.
I'm about 5m away from the AP. A tiny Ralink rt2800usb dongle in the
same machine gives solid performance (40...60 Mbs).
The experimental firmware didn't make a difference for me, except
it seems to fix /sys/kernel/debug/ieee80211/phy0/iwlwifi/iwlmvm/fw_rx_stats
(previously all values were 0, but I don't know how to interpret them).
> For what it's worth - on both this experimental firmware and previous
> revisions, I can do > 200Mbps to the public internet (on my office internet
> connection), with 13 other users currently on the same AP as me (the APs are
> Cisco 3702i's managed by a Cisco 2504.) This is (obviously) on the 5ghz
> band, with VHT enabled. I've had issues with occasional 'dropouts' with this
> card, where traffic will randomly stop passing and/or ping times will jump
> to > 10s; sometimes it clears up on its own, somethings I have to
> re-associate to the AP to get it to stop. I'm hoping the new firmware and/or
> the newest Cisco WLC build fixes it; testing now.
>
> Borislav - you didn't mention what firmware revision you are running; if you
> aren't on one of the -9 builds, I'd highly recommend moving to it.
>
> I do have the following set in
> /etc/modprobe.d/iwlwifi-disable-powersave.conf:
> "options iwlwifi power_save=0 bt_coex_active=0"
I also tried iwlmvm option power_scheme=1 (Continuously Active Mode),
11n_disable=1, and "iw dev wlan0 set power_save off". It stays slow,
and typing in ssh is still somewhat laggy.
> Once I move to 3.16, I'll probably try turning both of those features back
> on, and see what happens.
>
> Here's a speedtest result; I've gotten better, but this is representative of
> my average speeds:
> http://www.speedtest.net/my-result/3655207245
>
> In any case - this card _is_ capable of decent speeds for sure!
The question is how to go about testing. ISTR the Windows 8 driver
didn't have any issue and gave decent speed, but I don't have the
time to restore the Windows image for testing now. If Windows
and Linux use the same firmware I guess a comparison would
reveal if it's a driver of firmware issue.
Otherwise I'm hoping Intel guys could provide some guidance for
testing so we could find the root cause.
Johannes
On 07/29/2014 02:28 AM, Borislav Boev wrote:
> Hello, I am reporting this bug, though it is well known for years. In general I've experienced poor wifi performance with all Intel cards, starting from 3945abg, 4965agn and now Intel 7260 Dual-Band AC.
> The old cards were on a ThinkPad T61 system, 7260 is on a brand new ThinkPad Edge E540.
>
> So here's the problem:
> Wireless conection is slow, there's a unusual ping while browsing web pages (tried with all browsers - no luck), and sometimes drop-outs in connection.
> There's a partial fix by disabling Wireless "N" mode, but this is irrational, because it limits the network to only 54mbits speed.
>
> Details about the system:
> Hardware:
> ThinkPad E540
> CPU: Intel core i5-4200m (Haswell)
> Wireless adapter: Intel 7260 Dual-Band AC
>
> Software:
> PCLinuxOS 2014.07
> Kernel: 3.15.5 (though it was the same with old kernels)
> Kernel module: iwlwifi
You should try the experimental firmware file from
https://git.kernel.org/cgit/linux/kernel/git/egrumbach/linux-firmware.git/plain/iwlwifi-7260-9.ucode?h=Core6.
I am currently testing it and getting throughput of 60 Mbps on the 2.4 GHz band,
and 90 Mbps at 5 GHz.
Larry
On Tue, 29 Jul 2014, Larry Finger wrote:
> You should try the experimental firmware file from
> https://git.kernel.org/cgit/linux/kernel/git/egrumbach/linux-firmware.git/plain/iwlwifi-7260-9.ucode?h=Core6.
>
> I am currently testing it and getting throughput of 60 Mbps on the 2.4 GHz band,
> and 90 Mbps at 5 GHz.
For what it's worth - on both this experimental firmware and previous
revisions, I can do > 200Mbps to the public internet (on my office
internet connection), with 13 other users currently on the same AP as me
(the APs are Cisco 3702i's managed by a Cisco 2504.) This is (obviously)
on the 5ghz band, with VHT enabled. I've had issues with occasional
'dropouts' with this card, where traffic will randomly stop passing and/or
ping times will jump to > 10s; sometimes it clears up on its own,
somethings I have to re-associate to the AP to get it to stop. I'm hoping
the new firmware and/or the newest Cisco WLC build fixes it; testing now.
Borislav - you didn't mention what firmware revision you are running; if
you aren't on one of the -9 builds, I'd highly recommend moving to it.
I do have the following set in
/etc/modprobe.d/iwlwifi-disable-powersave.conf:
"options iwlwifi power_save=0 bt_coex_active=0"
Once I move to 3.16, I'll probably try turning both of those features back
on, and see what happens.
Here's a speedtest result; I've gotten better, but this is representative
of my average speeds:
http://www.speedtest.net/my-result/3655207245
In any case - this card _is_ capable of decent speeds for sure!
-Nate
On Wed, Jul 30, 2014 at 01:08:30PM +0200, Johannes Stezenbach wrote:
> The question is how to go about testing. ISTR the Windows 8 driver
> didn't have any issue and gave decent speed, but I don't have the
> time to restore the Windows image for testing now. If Windows
> and Linux use the same firmware I guess a comparison would
> reveal if it's a driver of firmware issue.
I had some spare time today to restore the Windows 8 image
and test (praise Clonezilla!). The test is with an AP which
(sometimes) sends broken beacons, so the result might not be viewed
as valid by some, but I'm just comparing performance of
different hardware and software under the same conditions.
https://bugzilla.kernel.org/show_bug.cgi?id=78101
Earlier today I could connect with Linux and the
Dual Band Wireless-AC 7260 gave me <50KB/s download
speed (fetching kernel tarball from PC connected via
1G ethernet to the AP). An rt2800 usb dongle on the
same machine gave me ~2.5MB/s. Currently Linux
cannot connect at all (see bug 78101) due to the
broken beacon.
With Windows 8 (Driver Version 16.10.0.5, Driver Date 1/28/2014)
I had no problems to connect, and got ~400KB/s download speed.
(I confirmed via capture on the PC that the beacons
were broken in the way described in bug 78101 during this test.)
The rt2800 usb dongle gave me 2.4MB/s in Windows.
I couldn't reboot the AP today to see if and how much
the performance improves when the beacons are not broken.
I'll do that when I have time. In the past, rebooting
the AP did fix the connection issue (good beacon for some time),
but did not turn the AC 7260 into a speed demon with the
Linux driver.
The AP only supports 2.4Ghz band 802.11n.
> Otherwise I'm hoping Intel guys could provide some guidance for
> testing so we could find the root cause.
So far Intel did not show much interest in the issue, but
a laptop user often has no control on the AP (travelling,
hotels, conferences, etc.). IMHO Intel would their users
a big favor if they spent some time optimizing for bad
network conditions. The broken AP on my side will be
replaced soon, but that's not the point of my complaint.
Thanks,
Johannes
On Sun, Aug 31, 2014 at 08:26:34PM +0300, Emmanuel Grumbach wrote:
> On Fri, Aug 29, 2014 at 5:50 PM, Johannes Stezenbach <[email protected]> wrote:
> >
> >> Otherwise I'm hoping Intel guys could provide some guidance for
> >> testing so we could find the root cause.
> >
> > So far Intel did not show much interest in the issue, but
> > a laptop user often has no control on the AP (travelling,
> > hotels, conferences, etc.). IMHO Intel would their users
> > a big favor if they spent some time optimizing for bad
> > network conditions. The broken AP on my side will be
> > replaced soon, but that's not the point of my complaint.
> >
> >
>
> So far - I was busy with other stuff - and will still be.
> After all, debugging issues we have with a completely broken AP isn't
> my top priority.
> And just for the record - can you share the progress you are having
> with the AP manufacturer? After all - this is the real bug - but you
> seem to complain about Intel more than the AP manufacturer... Strange
> isn't it?
I can't be bothered to talk to the AP manufacturer, this AP will
soon be history and current hardware revisions of it have a
different chipset. You see its beacon bug went unnoticed for
~2 years because every other device worked without issues with it.
But the AP isn't the issue, and I only mentioned bug 78101 in this
thread because I thought it would be unfair to talk about performance
comparisons and not mention it.
Wrt $subject, I heard so many people complain about Wifi performance
and connection issues (not limited to Intel hardware), and I had my
share of it, too. But usually a user doesn't have a clue
what is causing the problem, it could be the AP or "the internet"
or the microwave oven. And since the own laptop is working well at
home they don't suspect it is the laptop's Wifi adapter. I certainly
didn't, and only found out by accident because I have a bunch of USB
Wifi dongles for testing with some embedded systems and noticed
the embedded boards get much better performance than my two Thinkpads
(one AC 7260 and one Ultimate-N 6300 AGN).
Bottom line is that I would like to encourage others to do similar
comparisons against USB dongles in the 10€ price range.
Because apparently you seem to think there is no problem, because
people are not complaining, while I think there is the possibility
that people are not complaining because they are not aware who
to complain to. (I do recognize the posibility that noone
else has the same performance issues that I have, but how to find out?)
Regarding the Windows comparison, I did it mainly because I wanted to
see if I remembered correctly that it worked without any issues
(I went through the Windows update procedure that one is more or less
forced to do on first-time use, which means a few 100 MB download,
before saving and image and wiping the disk).
And I was somewhat suprised to see that even the Windows driver
has substandard performance compared to the cheap USB dongle.
Thanks,
Johannes
On Fri, Aug 29, 2014 at 5:50 PM, Johannes Stezenbach <[email protected]> wrote:
> On Wed, Jul 30, 2014 at 01:08:30PM +0200, Johannes Stezenbach wrote:
>> The question is how to go about testing. ISTR the Windows 8 driver
>> didn't have any issue and gave decent speed, but I don't have the
>> time to restore the Windows image for testing now. If Windows
>> and Linux use the same firmware I guess a comparison would
>> reveal if it's a driver of firmware issue.
>
> I had some spare time today to restore the Windows 8 image
> and test (praise Clonezilla!). The test is with an AP which
> (sometimes) sends broken beacons, so the result might not be viewed
> as valid by some, but I'm just comparing performance of
> different hardware and software under the same conditions.
> https://bugzilla.kernel.org/show_bug.cgi?id=78101
>
> Earlier today I could connect with Linux and the
> Dual Band Wireless-AC 7260 gave me <50KB/s download
> speed (fetching kernel tarball from PC connected via
> 1G ethernet to the AP). An rt2800 usb dongle on the
> same machine gave me ~2.5MB/s. Currently Linux
> cannot connect at all (see bug 78101) due to the
> broken beacon.
>
> With Windows 8 (Driver Version 16.10.0.5, Driver Date 1/28/2014)
> I had no problems to connect, and got ~400KB/s download speed.
> (I confirmed via capture on the PC that the beacons
> were broken in the way described in bug 78101 during this test.)
> The rt2800 usb dongle gave me 2.4MB/s in Windows.
>
> I couldn't reboot the AP today to see if and how much
> the performance improves when the beacons are not broken.
> I'll do that when I have time. In the past, rebooting
> the AP did fix the connection issue (good beacon for some time),
> but did not turn the AC 7260 into a speed demon with the
> Linux driver.
>
> The AP only supports 2.4Ghz band 802.11n.
>
>> Otherwise I'm hoping Intel guys could provide some guidance for
>> testing so we could find the root cause.
>
> So far Intel did not show much interest in the issue, but
> a laptop user often has no control on the AP (travelling,
> hotels, conferences, etc.). IMHO Intel would their users
> a big favor if they spent some time optimizing for bad
> network conditions. The broken AP on my side will be
> replaced soon, but that's not the point of my complaint.
>
>
So far - I was busy with other stuff - and will still be.
After all, debugging issues we have with a completely broken AP isn't
my top priority.
And just for the record - can you share the progress you are having
with the AP manufacturer? After all - this is the real bug - but you
seem to complain about Intel more than the AP manufacturer... Strange
isn't it?
On 08/03/2014 04:45 AM, Borislav Boev wrote:
> Okay, here's another 'workaround' (though I haven't tested it more than one day yet)
> Wifi performance seems to be better & stronger when I changed my network's broadcast channel (It was on auto select before, now I changed it to channel 11 - 2462mhz). I'm living in a populated area, there's a lot of wifi networks around, so that could be one of the problems, I bet most of them are on auto-channel and cheap routers aren't that good in selecting the best frequency, so there's a lot of interference between networks.
> The good thing so far is that I'm back on my wireless-N speeds. Will report if something goes wrong.
For what it is worth, I never let my router pick a channel using "auto". Most of
my neighbors are on channel 6, thus I use 1 or 11.
Larry