RE: [Vacm-general] Slow EMP handling

[San M. <net...@va...>]

Hey Ard,

The first thing I did when I got my cyclades was open it up :) Ted Tso,
the author of the Linux TTY/serial driver layer also had a look.. It
appears they use a *ton* of semi-standard UARTS which are connected to
an ISA bus on the system... fine for normal bursted serial flow.. but
really awfull for lots of single byte I/O which is what the EMP state
machine calls for rather heavily.... 

  This has been a common problem with serial concentrator vendors.... in
any event, playing in the kernel may be difficult as I understand their
serial driver itself is not open source.... other than that I agree that
it may be just a matter of software, however I just don't have the
bandwidth to work on yet another piece of serial hardware (I'm currently
working on preliminary IPMI 1.5 over LAN support)

If you do decide to take this on, let me know.. I'd be happy to provide
whatever EMP related assistance I can...
-san


-----Original Message-----
From: 'Ard van Breemen' [mailto:ar...@te...] 
Sent: Wednesday, July 25, 2001 5:13 AM
To: San Mehat
Cc: sa...@va...; vac...@li...
Subject: Re: [Vacm-general] Slow EMP handling

On Tue, Jul 24, 2001 at 11:29:06AM -0700, San Mehat wrote:
> We had done some initial investigating into the cyclades product to
see
> if this exact thing could be done... basically it appears that the box
> is too underpowered to drive all the ports with the EMP state
machine..
Hmmm, that is weird: I guess that with a 50MHz ppc, it is very easy to
control 32 serial ports.
> We noticed that the machine was so underpowered that it was taking a
few
> seconds for a connection to be acknowledged as closed by the box.. so
> the port that was being listened() on was showing a connection *well*
> after shutdown() was called on the socket.... not good...
Well, time to look at the source code then.... Hmmm, hands start
itching..
It sounds like somebody has programmed the box to generate interrupts
for every port.
This results on 19.2k (the emp speed) in 1920 interrupts per port per
second, that is about 61440 interrupts... Even a pentium will not be
able to handle that.
> The cyclades box looks to be a good solution for serial console
> redirection.. but it just doesn't look like it has the horsepower for
> managing the EMP state machine....
Just a question of software....
Like in the old days. Then they did the same, just with less hardware
:).
> -san

> Nope, all of the software that runs on the Cyclades TS1000 (16 port)
and
> TS2000 (32 port) term servers is open source. The main piece of
software
> that manages the ports is portslave or pslave. If you are curious the
> flash image of exactly what is on them is available for download from
> the Cyclades ftp site.
Yes! Cool!
Now I need the Cyclades itself, and unfortunately we already have serial
console systems, so I cannot request one for my work :(...

-- 
<ar...@te...> Telegraaf Elektronische Media  http://wwwijzer.nl
http://leerquoten.monster.org/ http://www.faqs.org/rfcs/rfc1855.html 
Let your government know you value your freedom. Sign the petition:
http://petition.eurolinux.org/