From: Matthew Trent <t3rmin@gm...> - 2005-01-02 22:33:06
I'm hoping somebody can give me a better understanding of how things
work with power management enabled and ndiswrapper. I read through the
Wiki, and it said ndiswrapper doesn't care about power management
settings (period, timeout...), just that it's turned on or off (I do a
iwconfig wlan0 power on). Any idea why this is? Does the Windows
driver have that stuff hard coded or has configurability of power
management not been implemented yet, or what?
Also, with power management turned on, I see very erratic latencies. I
can start a ping with power management off, and all the times are very
close together. When I turn on power management, though, the "jitter"
is very high, some pings lower, some higher, and in general latency is
quite a bit higher. Ssh or X over wireless with power management
enabled is almost unusable, but it's fine with power management off.
I realize power management comes at the cost of signal strength or
whatever, but I didn't expect it to be quite so noticeable (and I
figured it would probably adapt to usage, like CPU frequency scaling).
Is this correct behavior, or is something wrong? I have an Inprocomm
IPN 2220 and I'm using ndiswrapper 1.0rc1.
Thanks for ndiswrapper, by the way. I appreciate all the hard work.
On Sun, Jan 02, 2005 at 02:32:55PM -0800, Matthew Trent wrote:
> I realize power management comes at the cost of signal strength or
> whatever, but I didn't expect it to be quite so noticeable (and I
> figured it would probably adapt to usage, like CPU frequency
> scaling). Is this correct behavior, or is something wrong? I have an
> Inprocomm IPN 2220 and I'm using ndiswrapper 1.0rc1.
As far as I know, power management is typically going to affect
latency the most, since it usually involves reducing receptiveness to
incoming data. Under high load, I'm guessing it'll up its
responsiveness and power usage to match, but one tiny packet per
second and its reply isn't really high load.
It's been my experience that most hardware these days seems optimised
for throughput (i.e. web browsing) rather than latency. Take a look
at home broadband for an example of dreadful latency under high load
(due to excessive buffering for throughput).
Byte-at-a-time user-visible protocols haven't ever really been a part
of the Windows world, and these are Windows drivers on Windows-
oriented hardware. :)
That being said, this is mostly speculation based on what I've seen,
and should be taken as such.