Comments on Port
Steve Bush
Why Cygwin?
The port is a great thing for development for somebody like myself who
likes
to meld pc and Linux into one workstation.
I don't know if the cygwin port will be useful in a heavy *production*
environment. It doesn't have commercial support and performance under
Cygwin
might be an problem since I did a test and the file speed was nothing
like
as good as native Linux. This issue may or may not have a solution. It
remains to be seen. I believe reliability under Cygwin is fine.
Not a real Fork:
I don't consider it a proper fork that requires strong independent
support.
I consider it a minor change for a different Linux distro. Since the
port
required fairly limited changes, I or anybody who knows the Linux diff
tools
can maintain it up to the latest release quite easily. Cygwin is
reputed to
be a very complete POSIX platform except for fork and select which
don't
exist under Win32.
You can find all of the changes I made by typing something like grep -r
-i
cygwin * while at the cygwin prompt in the openqm folder.
Installation:
The steps to install together with screen shots of the basic Cygwin
install
are in the download file at
http://www.neosys.com/software/openqmcygwin.zip
It is quite possible that I left out critical steps at vital points etc
and
would be happy to step anybody through it in any messenger/chat tool of
your
choice.
Cygwin brings all blessings and curses of Linux to a PC prompt. If you
don't
have any Linux experience then don't expect Cygwin to be easy. It does
have
a really nice and simple pick and choose installation and upgrading
from
web, however you rapidly get into the typical editing of Linux
configuration
files for anything else. From the download sites offered I always use
http://mirrors.kernel.org near the bottom of the list.
Cheers,
Steve Bush
NEOSYS
Martin Phillips
[Sent to both groups]
Much as it is very brave of you to do a Cygwin port, I am not convinced of
its value. We did a Cygwin port ourselves long ago and dismissed it as
useless.
Just for fun, I installed your port on a system here and ran one of our file
system thrashing tests. This is designed to sustain a very high update rate
on a set of small files so that it is limited by the in-memory processing of
QM and the operating system, not the disk subsystem.
On your Cygwin port, we sustained an average transaction rate of 52 per
second. Using the native Windows implementation ran at 18360 per second.
Need I say more?
Martin Phillips, Ladybridge Systems
Steve Bush
Martin Phillips wrote:
> Hi Steve,
>
> [Sent to both groups]
>
> Much as it is very brave of you to do a Cygwin port, I am not convinced of
> its value. We did a Cygwin port ourselves long ago and dismissed it as
> useless.
Hardly brave. It was trivial as anyone can see from the IFDEF/IFNDEF
CYGWIN lines which mean that the port will happily compile under native
Linux or Cygwin without any changes.
Since this CYGWIN port doesnt compete with your PC product on
performance perhaps you could just include those IFDEF CYGWINS into
your core product in order to better support transparent compilation
under more Linux distro's ie including CYGWIN. As Stallman points out
often, Linux is actual GNU/Linux. In this sense Cygwin can be
considered GNU/Windows. I dont seek to prejudice your decision on the
matter of course.
>
> Just for fun, I installed your port on a system here and ran one of our file
> system thrashing tests. This is designed to sustain a very high update rate
> on a set of small files so that it is limited by the in-memory processing of
> QM and the operating system, not the disk subsystem.
>
Yes I can see your interest in the matter.
> On your Cygwin port, we sustained an average transaction rate of 52 per
> second. Using the native Windows implementation ran at 18360 per second.
> Need I say more?
>
I did a memory speed test which came out of my pushing the discussion
of OO in cdp newsgroup. It was a random 50,000 very small record hash
writes and took 600ms under Linux and the same on PC PersonalQM. It
took 60 seconds under Cygwin which is 100 times slower not the 4000
times difference. Maybe if I look into it I will find solutions. After
all CYGWIN is a very well supported and loved environment. I cant
believe it can have such bad performance.
Not wise to challenge me unless you are sure its a death march. :-)
Anyway, I hope the put up or shut up guys are happy now that I am not
just a mouth.
While the value of the openqmcygwin port for hacking/development of
GPLQM under win32 remains I agree that the prognosis for production
systems under CYGWIN is bad at this point in time. This leaves me with
my original problem of not having an open source version of QM under
win32 environment where all my clients are. For me this is a blocker as
discussed in the past.
Cheers,
Steve