tow devices share one interrupt

Norman Feske norman.feske at ...1...
Sat May 18 15:17:25 CEST 2013


Hello,

> finally,i combine the tow device into one,make the nitpicker use 
> input service route to the driver.it worked. but i find the android
> based mircrokernel is slower than native android without
> microkernel.is it because of the comunication of IPC? in genode,there
> is a lot of RPC to make it become server and client mode.what's the
> relationship between RPC and IPC?is that the reson that android
> become slower?

on Genode, each RPC corresponds to an IPC call.

There is no general answer on how much the inter-process communication
affects application performance. We always try to design the protocols
between components in a way that IPC is used in a sensible way. For
example, to communicate networking packets among components a Genode
system, we don't use one IPC per packet but shared-memory buffers
combined with asynchronous notifications. So many packet are passed from
one component to another at once.

Empirically evidenced by our experiences with optimizing application
performance in various scenarios, the speed of the IPC mechanism as
provided by the kernel was hardly ever be a dominating factor.

I suppose that the visibly slow performance of L4Android compared to
native Android can be mainly attributed to the different ways of how
graphics stack works. On native Android, Android employs the GPU for
rendering graphics, which is not only much faster than CPU-based
rendering, but it also offloads work from the CPU. In contrast,
L4Android processes all graphics rendering on the CPU an leaves the GPU
unused.

Regards
Norman

-- 
Dr.-Ing. Norman Feske
Genode Labs

http://www.genode-labs.com · http://genode.org

Genode Labs GmbH · Amtsgericht Dresden · HRB 28424 · Sitz Dresden
Geschäftsführer: Dr.-Ing. Norman Feske, Christian Helmuth




More information about the users mailing list