Hi, On Wed, Aug 29, 2012 at 10:00:30AM +0200, Thomas Knauth wrote: > >> Ah, yes, that worked. Except for the fact that the usb3380 does not > >> support streams :( I just found that out browsing the device's data > >> book ... The syslog also contains warnings/errors mentioning streams > >> ... so much for that. > > > > > > The driver does not support usb3 streams or the hardware? This is _one_ > > of the fancy USB3 things so I hope it is just the driver that is > > missing support for it. > > As its mentioned in the data book, I assume the HW itself misses > support for streams. > > > If you try to bump performance for tcm driver there is one thing you > > could do: Currently there is one request which waits for the command. > > In theory multiple requests could be queued and received. This should > > improve the performance to some degree because there is less waiting > > for the next command. > > I'm afraid my kernel/driver hacking skills are not evolved enough to > do that. As a an alternative to BOT/UAS I'm thinking about some > rudimentary support to transfer blobs of data over USB. Don't know if > this will eventually be faster without the support of streams though. you could just make a very low overhead function (no packet processing whatsoever) and just start a giant 1MiB transfer on each direction. That'll do for throughput measurement, I guess ;-) -- balbi
Attachment:
signature.asc
Description: Digital signature