On Tue, Mar 20, 2012 at 07:27:02PM +0200, Grazvydas Ignotas wrote:
On Tue, Mar 20, 2012 at 7:04 PM, Mark Brown
On Tue, Mar 20, 2012 at 06:15:34PM +0200, Grazvydas Ignotas wrote:
I wouldn't really call them broken, it's enough to set period size to 512 with smaller start_threshold (something like 50ms) to have problems, those parameters are perfectly valid for a program trying to achieve low latency.
If they can't cope with the parameters they've set I'd call them broken, they should've asked for more sensible parameters.
How is the program supposed to know those parameters are invalid for this hardware? It could maybe detect underflows and increase period until underflows stop, but there might be other reasons for underflows like high system load. Or do you mean setting up some period size and doing writes of that period size is not valid thing to do? Currently, no matter how fast the writes come, there is an underflow after first write in these conditions.
In that case why is the fix specific to this application and not a generic one? If we're going to underflow no matter what then why make the workaround custom?