loadtools with high baudrate give only marginal performance improvement

[Serg l]

Hi Mychaela,

Thank you for the feedback. I was using it with "classic" Slackware and it
was really annoying that I was getting this significant slowdown on Ubuntu.

-Serg

On Fri, Jul 5, 2019 at 3:07 PM Mychaela Falconia <
mychaela.falconia at gmail.com> wrote:

> Serg wrote:
>
> > I was running loadtools on Ubuntu with FTDI USB to UART interface and
> > noticed that increasing baudrate was giving just marginal load speed
> > improvement. Quick analysis revealed that select() in the S-record echo
> was
> > signaled by the kernel with 10ms delay. I tried multiple baudrates and
> > confirmed that that was pretty much constant.
>
> You must be seeing this behavior because you are running a too-new
> Linux kernel version - it does not happen with more classic kernels.
> I had implemented high baud rate support in loadtools back in 2013 and
> have been enjoying the faster image transfers ever since then without
> any problems.
>
> > Since we use Linux specific libserial now, I added one more kernel tweak,
> > which may be not reproducible on other platforms.
>
> Because the problem only happens with too-new Linux kernel versions,
> it is a Linux-specific problem, hence there is no problem with
> addressing it only in the Linux-specific version of libserial.
>
> > Maybe there is a better
> > place where we can set this flag, so here is a proof of concept.
>
> I will need to study the Linux kernel source and see exactly what this
> ASYNC_LOW_LATENCY flag does before I can incorporate this change into
> the official freecalypso-tools tree.  I will try to find some time,
> but I am not putting a high priority on it.
>
> > This makes an image load lighting fast compared to what I was getting
> > before. So far no major problems observed.
>
> Hmm, transferring a ~2 MiB (in the case of firmwares) or 8 MiB (in the
> case of forensic flash dumps) image over 812500 baud with each byte
> sent as two hex characters is not what I would call lightning fast -
> it is certainly tolerable, much more tolerable than doing the same
> over 115200 baud, but I would not call it lightning fast by any means.
>
> Oh wait, you said it's lightning fast *compared to what you were
> getting before*.  Does that statement mean that you had never
> experienced how well it works and has always worked with an older
> host system?
>
> M~
> _______________________________________________
> Community mailing list
> Community at freecalypso.org
> https://www.freecalypso.org/mailman/listinfo/community
>