[2372] in linux-net channel archive
Re: Internet Provider?
daemon@ATHENA.MIT.EDU (Theodore Ts'o)
Fri Apr 5 14:52:39 1996
Date: Fri, 5 Apr 1996 14:29:12 -0500
From: "Theodore Ts'o" <tytso@mit.edu>
To: Alan Cox <alan@cymru.net>
Cc: dledford@dialnet.net, hays@satcom.whit.org, HADES@lapro4.unisinos.tche.br,
linux-net@vger.rutgers.edu
In-Reply-To: Alan Cox's message of Fri, 5 Apr 1996 11:33:22 +0100 (BST),
<199604051033.LAA01663@snowcrash.cymru.net>
From: Alan Cox <alan@cymru.net>
Date: Fri, 5 Apr 1996 11:33:22 +0100 (BST)
> I personally recommend the Comtrol RocketPort serial cards. They use a
> 1023 byte recieve FIFO and a 255 byte transmit FIFO (according to the
> driver header files). I have gotten great performance out of them
> myself, even under loads such as 48 simultaneous PPP connections running
> on a Pentium 100. I have hacked a little code into the driver to give me
> some timing stats on CPU time used per instance of the polling routine as
> well as a high and low watermark for the polling routine over a ten
> minute time frame. If anyone is interested in the stats, let me know.
Until we know the comtrol drivers work 100% for all uses I'd be wary of
recommending them. Hopefully the new driver will do the trick. Cyclades
boards are rock solid and far better than 16550's because they help out
a lot (things like flow control are software driven on a 16550 which is
bad). The rocketports are very very fast and very efficient.
I have been working with various ISP/BBS admins to make the comtrol
drivers more stable in the face of increasing loads. The problem was
that when I developed it, I didn't have the testing facility to test it
with large numbers of ports. Alan Cox, Doug Ledford, and Jon Lewis have
all been very helpful in reporting both bugs and fixes to work out these
sorts of problems.
At this point, 1.09 seems to be solid pretty solid for ~24 ports, and
I've fixed two more bugs in 1.10 which seem to make it work well with
32-48 ports in use. 1.10 may be solid with more ports in use, but only
time and experimentation will tell. I do have one set of changes which
is not yet in 1.10 to fix what may be a potential race condition bug.
I'm waiting to hear from Doug to see if that fixes the last of his
problems. (Which is a relatively rare, hard lockup w/o a kernel oops
message, and it's not clear whether or not it's really a rocketport
driver problem or not, or in some other part of the 1.2 kernel. We have
a theoretical patch which may help, and it's currently in testing.)
My recommendation is that if you're using 32 or fewer ports, the 1.10
Rocketport driver should work just fine. With more ports than that,
things will probably be stable, but more experimentation is clearly
needed. I am confident that we should be able to fix any future
problems that crop out without too much difficulty, especially
considering the help and cooperation I've gotten from people like Alan,
Jon, and Doug. Thanks, guys!!
- Ted
