I meddled around with this last week via the Fedora ARM list, it was simple to reproduce that pinging the Panda Ethernet from another machine on the local network with -i0.001 yielded an effective 50% increase in USB storage throughput at the Panda compared to it being left idle.
Neither RUNTIME_PM (suggested by Felipe Balbi at TI) nor USB_SUSPEND were to blame since the issue stayed the same with those deconfigured.
It seems the issue may be to do with the Panda's hub / ethernet chip idle / wakeup characteristics but since there are no hub registers in there settable from Panda side AFAIK (there are some settable in the missing EEPROM that can configure the hub / Ethernet bridge) I am out of ideas to follow it up.
I meddled around with this last week via the Fedora ARM list, it was simple to reproduce that pinging the Panda Ethernet from another machine on the local network with -i0.001 yielded an effective 50% increase in USB storage throughput at the Panda compared to it being left idle.
Neither RUNTIME_PM (suggested by Felipe Balbi at TI) nor USB_SUSPEND were to blame since the issue stayed the same with those deconfigured.
It seems the issue may be to do with the Panda's hub / ethernet chip idle / wakeup characteristics but since there are no hub registers in there settable from Panda side AFAIK (there are some settable in the missing EEPROM that can configure the hub / Ethernet bridge) I am out of ideas to follow it up.