> > Trying to save to a SATA harddrive as fast as possible and I need
> > more speed - currently I am using a USB external drive.
> > My drive is connected on the musb bus if I connect it on an EHCI
> > bus does that make a significant speed difference?
> > My understanding - probably flawed - is that EHCI bus will be
> > faster than musb?
> > Alternatively is there a better way to get fast write speeds
> > instead of using USB on the Omap3530 platform?
> > I am guessing that USB is the only real way of doing this, even
> > if we use SATA as that will be going via USB anyway.
> Have you looked into using SDIO/SD/MMC? It's not as theoretically
> fast as USB 2.0 but it's less likely to be a shared bus.
Yeah - MMC is slightly faster, but not significantly.
> What exactly do you need 10s of MB/sec of throughput to?
We are writing data from a camera.
What kind of camera needs 10s of MB/sec throughput?!?!
My home camcorder does 1080i HD at like 18 Mb/sec (that's bits, with a
little 'b' not bytes with a big 'B'). Even over the air HD is usually
under 20 Mb/sec for 1080i. BluRay is _ONLY_ 36 Mb/sec. All those are
bits per second numbers. You're talking tens of Mega bytes per
second? From one camera?
Are you really pushing that kind of data around that you're saturating
USB 2.0? If so, compression is going to be your friend. Compress
before you send to disk.
Are you doing 4K video uncompressed at 60fps or something?
Any other video stream boggles me why USB 2.0 wouldn't keep up, even
if the bus had other operations on-going.
My initial testing used DD but showed massively different numbers
depending on the settings I use - so instead of DD I have focussed on
frame rate saved as the speed indicator.
So at this stage I need double the bandwidth.
> Does it have to be a spinning rust hard disk?
Yes, we need 2TB or more.
> Would a ramdisk work?
I was thinking of trying some sort of caching system, but at the end
of the day it is the drive that is the bottle neck.
> Would a USB connected SSD give better throughput?
Probably not - but I haven't tested it - I am thinking that the
actual USB bus rather than the drive is the bottle neck.
Have you tested this assumption?