I’m having a look at UMDF, and one of the caveats in using such framework is
obviously latency.
In one of the presentations I found (winhec 2006, i think) it’s clearly
stated that “UMDF may increase latency, but throughput remains high”.
Are there any numbers or estimates for such latencies for, let’s say, an i/o
request (i.e. from the DeviceIoControl call to when the DispatchIoCtl
callback in the UMDF driver gets called)?
I’m not interested in official numbers, just a rough idea…
GV
“Gianluca Varenni” wrote in message news:xxxxx@ntdev…
> I’m having a look at UMDF, and one of the caveats in using such framework is obviously latency.
>
> In one of the presentations I found (winhec 2006, i think) it’s clearly stated that “UMDF may increase latency, but throughput
> remains high”.
>
> Are there any numbers or estimates for such latencies for, let’s say, an i/o request (i.e. from the DeviceIoControl call to
> when the DispatchIoCtl callback in the UMDF driver gets called)?
>
> I’m not interested in official numbers, just a rough idea…
Roughly estimating, this might be couple of task switching times:
the app requesting i/o to the UMDF host process, and back.
And then add 3-4 kernel call times.
–PA