On Mon, 9 Jan 2012, Felipe Balbi wrote: > > Okay, so there are some hardware setups that won't go as fast as they > > might. The lossage isn't all that great, because even at high speed > > UAS isn't that much faster than BOT. > > > > (Although, I admit, it would be nice to have some benchmark figures to > > back up this claim...) > > When I measured our (I mean Linux's) implementation of UASP and > g_mass_storage were behaving quite the same throghput-wise. But then > again, device and host were quite powerful machines (quadcore with 16GiB > RAM). I guess when CPU(s) is(are) busy, that memcpy which we save when > using VFS's SG list will pay off. Well, what matters more is the difference when using common USB-3 disk hardware, not a Linux gadget. And even then, the saving doesn't come so much from buffer copying. A non-SG BOT transfer shouldn't do any more copies than a UAS transfer. The saving comes from not having to allocate and submit a whole bunch of URBs for each transfer, from not using a separate kernel thread, and from having more endpoints (which allows greater concurrency of transfers). Alan Stern -- To unsubscribe from this list: send the line "unsubscribe linux-usb" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
