Re: 2.6.38: rt2800usb: high latency (1000ms)?

[Justin Piszcz]

On Wed, 13 Apr 2011, Daniel Halperin wrote:

> Ugh! Couldn't you configure the stupid mailing list filter to only
> drop rich text mails that have [PATCH or [RFC or [RFT in the subject,
> e.g.? Original mail below.
>
> (sorry for the resend, Justin and Ivo).
>
> Dan
>
> On Wed, Apr 13, 2011 at 9:23 AM, Daniel Halperin
> <dhalperi@xxxxxxxxxxxxxxxxx> wrote:
> > On Wed, Apr 13, 2011 at 9:01 AM, Justin Piszcz <jpiszcz@xxxxxxxxxxxxxxx> wrote:
> > > > Can you try disabling powersaving?
> > > > iwconfig wlan0 power off
> > > Wow! That was it, now its interactive again.
> > >
> > > 64 bytes from wireless-host (192.168.1.2): icmp_req=1 ttl=64 time=0.581 ms
> > > 64 bytes from wireless-host (192.168.1.2): icmp_req=2 ttl=64 time=0.647 ms
> > >
> > > Thanks, so its recommend to keep this off then, can that be set as the driver default?
> >
> > That's a bad idea, what you're seeing is likely completely a red herring. Wi-Fi power saving mode saves energy by putting the RF hardware most of the time. This is a Good Thing.
> > Power save is designed to kick in only when the load is very slow; the client's network stack should automatically disable power save during times of high load, e.g., during a download or a VoIP call. You should run some tests to see if this is actually occurring; there may be a bug in the rtl implementation.
> > The real problem here is that your ping test has invalid assumptions. One packet/second is not enough load to disable Wi-Fi power save, so you should not see interactive ping times. Again, this is a good thing and something I doubt you want to disable! (Assuming that RTL's implementation isn't fundamentally broken elsewhere) It certainly should NOT be the driver default.
> > Try again with sudo ping -i 0.01 -s 1400 (e.g., to ping with large packets every 10 ms); this should probably trigger the logic that automatically disables the power saving mode. Or try pinging while doing a 1 Mbit UDP transfer (e.g., with iperf).
> > Dan

Hi,

When powersave is enabled, it is very jumpy, I've used satellite comms before
and (~600ms-1200ms was more smooth) as it did not jump around as much.  The
application is just a standalone desktop with minimal activity for the majority
of the time, maybe thats why..

With powersave disabled, I now see 0% packet loss (802.11n) and low ping
times, this looks like the proper solution for the wireless USB device I
am using.  By the way, is it possible/are there wireless USB devices out there
that support wake on wireless lan (WOWL?

Your ping command with power off:
1408 bytes from server (192.168.1.2): icmp_req=539 ttl=64 time=1.07 ms
1408 bytes from server (192.168.1.2): icmp_req=540 ttl=64 time=1.31 ms
1408 bytes from server (192.168.1.2): icmp_req=541 ttl=64 time=1.07 ms
1408 bytes from server (192.168.1.2): icmp_req=542 ttl=64 time=1.26 ms

Your ping command with power on:
1408 bytes from server (192.168.1.2): icmp_req=649 ttl=64 time=1.80 ms
1408 bytes from server (192.168.1.2): icmp_req=650 ttl=64 time=1.85 ms
1408 bytes from server (192.168.1.2): icmp_req=651 ttl=64 time=2.86 ms
1408 bytes from server (192.168.1.2): icmp_req=652 ttl=64 time=1.46 ms

You are correct, if there is a lot of traffic, its good, but if the system
is relatively idle and all that's going on is an SSH session, there is horrible
latency.

So the fix/workaround, for Debian-based distributions (if you come across 
this problem):

File: /etc/network/interfaces

-- snip --

# Wireless
auto wlan0
iface wlan0 inet dhcp
  wpa-ssid your-ssid
  wpa-psk your-private-key
  post-up /sbin/iwconfig wlan0 power off

-- snip --

Justin.