glenm@mako.UUCP (01/03/85)
For a typical mix of shell/edit/compile jobs on an 11/780, does anyone know how UNIX compares with VMS in terms of performance? By performance is meant response time and throughput time. I realize that this is an involved question, and that there are lots of other arguments in favor of UNIX. This is being done for a friend whose site is stuck with VMS. I would also like to know, if one OS is significantly better than the other, what the reasons are. Thanks. Glen McCluskey ..tektronix!mako!glenm
rcb@rti-sel.UUCP (Randy Buckland) (01/04/85)
> > I realize that this is an involved question, and that there are > lots of other arguments in favor of UNIX. This is being done for > a friend whose site is stuck with VMS. > Then this question is not going to help your friend. I don't have any exact figures, but from just general experience on a 750 running BSD 4.2 and a 750 running VMS, the VMS machine is significantly faster. I would guess that it is about 1.4 to 1.8 times faster in general execution speed. This is probably attributable to the faster file system and I/O system. Randy Buckland Research Triangle Institute ...!mcnc!rti-sel!rcb
Doug Gwyn (VLD/VMB) <gwyn@Brl-Vld.ARPA> (01/04/85)
Please! Let's not start another VMS-vs-UNIX debate!
The original question this time is about RELATIVE PERFORMANCE,
which implies that a DEC-supported VAX configuration must be
assumed. I hope everyone who decides to address this issue
will realize that narrowly-defined "performance" should seldom
be the determining factor in selecting an operating system.
If respondents will stick to the technical issue of performance
this topic could be interesting.
My contribution to the issue at hand will be limited to quoting
AT&T performance figures for UNIX System V Release 2.0 for the
VAX-11/780 with and without KMC11-B I/O processor assist for the
DZ11 terminal multiplexor. The VAX-11/750 performance is roughly
2/3 of that of the VAX-11/780:
C library functions (times in msec):
strcmp of 64 bytes: .106
atof: .455
fprintf of double: .639
putc: .013
getc: .013
fputs of 64 bytes: .248
fgets of 64 bytes: .250
fwrite of 1K bytes: 1.50
fread of 1K bytes: 2.34
Kernel functions (times in msec):
getpid: .180
context switch: .750
chdir("."): 1.27
open("file",0): 2.50
pipe 1K bytes: 3.96
read 1 byte (cache): .850
read 1K bytes (cache): 1.71
access disk block: 3.1
read 4K bytes file: 20.7
fork/exit (8K data): 22.2
exec (8K bss): 19.4
exec (8K data): 27.7
exec (512 byte args): 64.1
Terminal output, 9600 baud, 20 ports (rates in K bytes/sec):
RAW mode:
Without KMC assist:
1-byte write(): 0.7
512-byte write(): 7
With KMC assist:
1-byte write(): 0.8
512-byte write(): 50
COOKED mode:
Without KMC assist:
1-byte write(): 0.8
512-byte write(): 2
With KMC assist:
1-byte write(): 0.7
512-byte write(): 50
Terminal input, 10 ports:
RAW mode:
Without KMC assist:
1-byte read(): 0.8
256-byte read(): 1.5
With KMC assist:
1-byte read(): 1.3
256-byte read(): 4
COOKED mode:
Without KMC assist:
1-byte read(): 0.7
256-byte read(): 1.2
With KMC assist:
1-byte read(): 1.0
256-byte read(): 2.5
Simulated system load using shell scripts, not interactive user
(throughput in jobs[not processes!]/hour):
1 script 7500
2 simultaneous scripts 10000
5 simultaneous scripts 10800
10 simultaneous scripts 10700
20 simultaneous scripts 10000
Summary: VAXes running AT&T UNIX System V Release 2.0 really should
be configured with KMC11-B assist for the DZ11s. 128 KMC/DZ ports
are advertised, and the quoted figures show that terminal I/O is
not a significant bottleneck even for this many simultaneous users.
The system response degradation under heavy load is quite decent,
such that 64 simultaneous users would not be at all unreasonable,
so long as they were willing to each have only 1/64th of the CPU
resources. The real problem is that 1/64th of a VAX-11/780 is not
much power; all operating systems for the VAX share this difficulty!
I hope that somebody posts similar measurements for 4.2BSD and for
VAX/VMS, so that meaningful comparisons can be made.