Subject: first attempt at LMBENCH run for SunOS-4 vs NetBSD on 3/260...
To: None <port-sun3@NetBSD.ORG>
From: Greg A. Woods <woods@most.weird.com>
List: port-sun3
Date: 06/07/1996 00:21:41
Here's the quick-format summary from a first hack at running lmbench-1.1
on two similar Sun 3/260's, one under SunOS-4, the other under NetBSD.

Most is my main server, running SunOS 4.1.1_U1, and was running normal
mail processing during the tests.  I only gave the memory tests 8 MB to
work with even though the machine has 48 MB.

Always is a test system running stock NetBSD-1.1.  It was running NFS
mounted partitions to access the lmbench programs, but was otherwise
untouched (except for syslogd, timed, cron, rwhod, etc.)

As you'll see the file reread on "always" didn't complete due to
overflowing the filesystem.  I'll run it again tonight now I've cleared
some space (no more games and esp. no more fortunes for this tiny disk!)
and killed the extra daemons.

I have no idea why yet, but lmbench (mhz) gets the clock speed very
wrong on these machines....  It claims ~9.85 MHz for the 3/260, and
about 6.32 MHz for my 3/60 (though the latter was run under a full
operating X11 environment).  Perhaps lmbench just doesn't like the
mc68k?  The code in mhz.c is quite simple, but I've not yet looked at
the compile output to see what's up.

most.weird.com:		3/260
			48 MB [2x16 + 2x8]
			sd0:  <Quantum PD 1050S cyl 2444 alt 2 hd 12 sec 70>
			SunOS Release 4.1.1_U1 (MOST) #1: Wed Jan 31 14:19:12 EST 1996
			/usr/5bin/cc -O

always.weird.com:	3/260
			32 MB [4x8]
			si0 targ 0 lun 0: <TOSHIBA, MK234FB, 1.81> SCSI1 0/direct fixed
			sd0 at scsibus0: 101MB, 853 cyl, 7 head, 34 sec, 512 bytes/sec
			NetBSD 1.1 (GENERIC) #24: Thu Dec 14 13:52:07 EST 1995
			    gwr@venus:/home/gwr/work/src/sys/arch/sun3/compile/GENERIC
			/usr/bin/gcc -O


                    L M B E N C H  1 . 0   S U M M A R Y
                    ------------------------------------

            Processor, Processes - times in microseconds
            --------------------------------------------
Host                 OS  Mhz    Null    Null  Simple /bin/sh Mmap 2-proc 8-proc
                             Syscall Process Process Process  lat  ctxsw  ctxsw
--------- ------------- ---- ------- ------- ------- ------- ---- ------ ------
always       NetBSD 1.1   10     174     59K    393K    737K 3258    224    464
most      SunOS 4.1.1_U   10     170     35K    160K    257K 1480    183    653

            *Local* Communication latencies in microseconds
            -----------------------------------------------
Host                 OS  Pipe       UDP    RPC/     TCP    RPC/
                                            UDP             TCP
--------- ------------- ------- ------- ------- ------- -------
always       NetBSD 1.1    1339    2814    7266    3604    7879
most      SunOS 4.1.1_U    1260    2540    5185    2860    5549

            *Local* Communication bandwidths in megabytes/second
            ----------------------------------------------------
Host                 OS Pipe  TCP  File   Mmap  Bcopy  Bcopy  Mem   Mem
                                  reread reread (libc) (hand) read write
--------- ------------- ---- ---- ------ ------ ------ ------ ---- -----
always       NetBSD 1.1    1    1      0      6      3      4    7     9
most      SunOS 4.1.1_U    2    1      2      4      4      4    7     9

            Memory latencies in nanoseconds
            (WARNING - may not be correct, check graphs)
            --------------------------------------------
Host                 OS   Mhz  L1 $   L2 $    Main mem    Guesses
--------- -------------   ---  ----   ----    --------    -------
always       NetBSD 1.1     9   138    948         948    No L2 cache?
most      SunOS 4.1.1_U     9   137    997         977    No L2 cache?


So, I see lots of significantly higher latencies in NetBSD/sun3!  Most
scary of all is the pipe bandwidth -- this hints of some rather nasty
throughput problems, no?  Esp. considering the SunOS box was under full
multi-user online processing while running the benchmark.

Here's the same data from 'make percent':


                L M B E N C H  1 . 0   S U M M A R Y
                ------------------------------------

                  Comparison to best of the breed
                  -------------------------------

                (Best numbers are starred, i.e., *123)


        Processor, Processes - factor slower than the best
        --------------------------------------------------
Host                 OS  Mhz    Null    Null  Simple /bin/sh Mmap 2-proc 8-proc
                             Syscall Process Process Process  lat  ctxsw  ctxsw
--------- ------------- ---- ------- ------- ------- ------- ---- ------ ------
always       NetBSD 1.1   10     1.0     1.7     2.5     2.9  2.2    1.2   *464
most      SunOS 4.1.1_U   10    *170  *33.9K *156.3K *250.8K *1480   *183    1.4

        *Local* Communication latencies - factor slower than the best
        -------------------------------------------------------------
Host                 OS  Pipe       UDP    RPC/     TCP    RPC/
                                            UDP             TCP
--------- ------------- ------- ------- ------- ------- -------
always       NetBSD 1.1     1.1     1.1     1.4     1.3     1.4
most      SunOS 4.1.1_U   *1260   *2540   *5185   *2860   *5549

        *Local* Communication bandwidths - percentage of the best
        ---------------------------------------------------------
Host                 OS Pipe  TCP  File   Mmap  Bcopy  Bcopy  Mem   Mem
                                  reread reread (libc) (hand) read write
--------- ------------- ---- ---- ------ ------ ------ ------ ---- -----
always       NetBSD 1.1  82%   *0    20%     *5    76%     *4   *7    *9
most      SunOS 4.1.1_U   *1  71%     *2    73%     *3    94%  97%   93%

            Memory latencies in nanoseconds - factor slower than the best
                    (WARNING - may not be correct, check graphs)
            -------------------------------------------------------------
Host                 OS   Mhz  L1 $   L2 $    Main mem    Guesses
--------- -------------   ---  ----   ----    --------    -------
always       NetBSD 1.1     9   1.0    ???        *948    No L2 cache?
most      SunOS 4.1.1_U     9  *137    ???         1.0    No L2 cache?


Given it takes well over an hour to run, I'll schedule some downtime
next time I'm out of the house for a short period of time and try
running the benchmarks in single user mode on the SunOS box a couple of
times.

Anyone interested can obtain the raw results from me.

-- 
							Greg A. Woods

+1 416 443-1734			VE3TCP			robohack!woods
Planix, Inc. <woods@planix.com>; Secrets of the Weird <woods@weird.com>