Samba/VMS 2.2.7a performances

[COLLOT Jean-Yves]

Hi.

I am almost ready to give a new release of Samba 2.2.7a. It will be
available later this week, or at least at the beginning of next week. 

The main change is that I completely re-wrote the TDB stuff (processing of
the files who keep information about open sessions, locks on open files and
other things). From the beginning, I had a lot of problems with the original
code, due to the fact that it uses shared memory mapped files in a non
VMS-compatible way. I got either corrupted files, or poor performances. So I
eventually decided to do what I should have done from the beginning: forget
about the mapped files and use indexed RMS files.

Note that the drawback of the next release is that those of you who has
joined a NT domain (smbpasswd -r xxx -j ddd) will have to do it again after
installing the new release... 

The main purpose of this message is to give you some information about
performances of this new release.

I made some testing with Samba/VMS compared with Samba/Unix(tru64) and
native Windows networking. The machines I used where very similar for VMS
and Tru64, and a little less powerful for Windows. The VMS and Tru64 boxes
are DS20E 666mhz, with efficient disks (HSZ70 controller with mirrorsets of
RZ 4Mb disks).

I started with the perl script that creates, reads and deletes a lot of
files. The result is not so good. For 200 files, we get the following
figures, in seconds, for respectively Windows, Tru64 and VMS :

Create		9	10	46
Open/Close		4	3	4
Delete		3	5	21

Well, Samba/VMS is not very good at creating and deleting files, is it?
That's not so surprising, because VMS file system is not too good at that
either. The key point is that every changes in the VMS file system is
immediately written back to disk (Unix or Windows don't do that: they keep
everything in the file cache and actually write to disk later). 

However, in my opinion, creating hundreds of almost empty files and deleting
them immediately after is hardly typical of a PC client activity, so I made
some more testing, using "ptime", a windows utility that can be found on the
Net and gives the opportunity of measuring the time elapsed for running a
command.

I tried 3 actions :
1. Copy of a 8 Mbytes files from the local PC to the remote computer (Samba
or other)
2. Open with Notepad of a 4 Mbytes file
3. "dir" of a directory containing 300 files

Here are the results (again in seconds, with Windows, Tru64 Samba and VMS
Samba):

Copy 8Mo		3.9	8.0	5.5
dir 300 files	1.0	0.2	0.6/0.2
notepad 4Mb		4.0	5.8	3.0

Note that the 2 values for dir/VMS are for the first time and the subsequent
times, showing the effectiveness of the memory caching process.

For those tests, we can see that now, Samba/VMS is very similar, and
sometimes better than Samba/Unix or native Windows (remember that my Windows
testing box is less powerful than the tru64/VMS ones).

To sum up, I think (and hope) that the upcoming release of Samba/VMS will be
quite useable for standard users, most of the standard activity being done
in an amount of time similar to other platforms (with the exception, anyway,
of creating (copy to VMS) or deleting a great number of files, that remains
significantly slower on VMS).

I hope that you'll use and enjoy this new release.

JY 
-------------- next part --------------
HTML attachment scrubbed and removed