I have a server application ported from Windows to Solaris. This is a
backup/restore application, and the main thing it does is: it receives
files, stores them and send them back when a restore request is issued.
I have a Ultra-2 (2 CPU x 300 MHz) box with Solaris 8, and another Sun Blade
100 (1 CPU at 500 MHz) also with Solaris 8.
The problem is that the transfer speed from a Windows box to the Ultra-2 one
is ~2,7 MB/s while the transfer back from Solaris to Windows, goes 6 MB/s
and up. By doing the same thing between the same Windows machine and the Sun
Blade, the backup gives about the same speed like the restore (~6MB/s)
I completely suppressed the writing to the disk, and on the Ultra 2 I cannot
get more than 3 MB/s. a simple ftp gives ~ 9MB/s from Windows to Ultra 2.
Does anybody have any idea what goes wrong on the Ultra 2 machine? How could
I track this bottleneck? The application looks to be OK since on the other
machine the speed is as expected. I also suspected the network, but the ftp
is OK. Are there any relevant OS patches which could be installed on Sun
Blade only (which is the development computer) and affect the speed? Or is
just the hardware? Assuming such a scenario, why the restore goes much
The socket receive buffer (for backup) and the send buffer (for restore) are
set to 0x8400 (33792). This size was set some years ago, after playing with
different numbers (on Windows)
Any help would be greatly appreciated