[Search for users] [Overall Top Noters] [List of all Conferences] [Download this site]

Conference turris::digital_unix

Title:DIGITAL UNIX(FORMERLY KNOWN AS DEC OSF/1)
Notice:Welcome to the Digital UNIX Conference
Moderator:SMURF::DENHAM
Created:Thu Mar 16 1995
Last Modified:Fri Jun 06 1997
Last Successful Update:Fri Jun 06 1997
Number of topics:10068
Total number of notes:35879

8203.0. "Digital UNIX + Baan IV/Triton" by VAXRIO::LEO () Fri Dec 13 1996 12:23

T.RTitleUserPersonal
Name
DateLines
8203.1DCULTX::hanQuestions raise the doubtMon Dec 16 1996 04:4917
8203.2VAXRIO::LEOMon Dec 16 1996 06:0610
8203.3KITCHE::schottEric R. Schott USG Product ManagementTue Dec 17 1996 22:0742
8203.4KITCHE::schottEric R. Schott USG Product ManagementTue Dec 17 1996 22:1114
8203.5Will doDCULTX::hanQuestions raise the doubtWed Dec 18 1996 01:5811
8203.6SMURF::MENNERBorn stupidWed Dec 18 1996 08:221
8203.7VAXRIO::LEOMon Dec 30 1996 12:0020
8203.8Fifo-Stats related to the Baan testVAXRIO::LEOMon Jan 27 1997 15:37166
Hi,

	I'm transmiting the informations that Eric has asked for on  either 
8203.3 and 8203.4 notes.

******************************************************************************

As Eric suggested :

>> A) dbx -k /vmunix
>>    p fifo_stats (before)
  
   fifo_wbolts_tot          0
   fifo_wblots_deliv        0
   fifo_wbolts_force        0
   fifo_rbolts_tot        216
   fifo_rbolts_deliv       72
   fifo_rbolts_force       41
   fifo_found_newdata       0
   fifo_active_fifonodes  264

>> run test

>>    p fifo_stats (after)

   fifo_wbolts_tot          0
   fifo_wblots_deliv        0
   fifo_wbolts_force        0
   fifo_rbolts_tot     172226
   fifo_rbolts_deliv   172062   
   fifo_rbolts_force    86881
   fifo_found_newdata       0
   fifo_active_fifonodes  245

Results of this execution :

MRP : 14 minutes and 41 seconds 
CRP :  8 minutes and 23 seconds 

------------------------------------------------------------------------------

>> B) dbx -k /vmunix
>>    ass fifo_do_adaptive=0
>>    p fifo_stats (before)
  
   fifo_wbolts_tot          0
   fifo_wblots_deliv        0
   fifo_wbolts_force        0
   fifo_rbolts_tot        229
   fifo_rbolts_deliv       86
   fifo_rbolts_force       67
   fifo_found_newdata       0
   fifo_active_fifonodes  264

>> run test

>>    p fifo_stats (after)

   fifo_wbolts_tot          0
   fifo_wblots_deliv        0
   fifo_wbolts_force        0
   fifo_rbolts_tot      40196
   fifo_rbolts_deliv    40034   
   fifo_rbolts_force   188988
   fifo_found_newdata       0
   fifo_active_fifonodes  212

Results of this execution :

MRP :  2 minutes and 34 seconds 
CRP : 58 seconds 

-----------------------------------------------------------------------------

>> C) a third test :
>>    
>>    p fifo_stats (before)
  
   fifo_wbolts_tot          0
   fifo_wblots_deliv        0
   fifo_wbolts_force        0
   fifo_rbolts_tot        220
   fifo_rbolts_deliv       77
   fifo_rbolts_force       37
   fifo_found_newdata       0
   fifo_active_fifonodes  264
>>
>>
>> run test
>>
>>    p fifo_stats (after)

   fifo_wbolts_tot          0
   fifo_wblots_deliv        0
   fifo_wbolts_force        0
   fifo_rbolts_tot     172391
   fifo_rbolts_deliv   172246   
   fifo_rbolts_force    86782
   fifo_found_newdata       0
   fifo_active_fifonodes  233

Results of this execution :

MRP : 14 minutes and 40 seconds 
CRP :  8 minutes and 17 seconds

>>
>> changes /etc/sysconfigtab
>>
>>	vfs:
>>
>>		pipe-maxbuf-size = 131072
>>
>> reboot the system
>>
>> run test
>>
>> p fifo_stats (after vfs, aka pipe-maxbuf-size)
>>
   fifo_wbolts_tot          0
   fifo_wblots_deliv        0
   fifo_wbolts_force        0
   fifo_rbolts_tot     216698
   fifo_rbolts_deliv   214532   
   fifo_rbolts_force   279260
   fifo_found_newdata       0
   fifo_active_fifonodes  818

Results of this execution :

MRP : 14 minutes and 41 seconds 
CRP : 46 seconds

******************************************************************************

The complete chart follows bellow :


--------------------------------------------------------------------------------
                   	(A)     (A)   (B)     (B)     (C)    (C)         (C) 
                       before  after before  after   before after     after vfs
-------------------------------------------------------------------------------
fifo_wbolts_tot          0       0     0       0       0      0           0
fifo_wblots_deliv        0       0     0       0       0      0           0
fifo_wbolts_force        0       0     0       0       0      0           0
fifo_rbolts_tot        216  172226   229   40196     220 172391      216698
fifo_rbolts_deliv       72  172062    86   40034      77 172246      214532 
fifo_rbolts_force       41   86881    67  188988      37  86782      279260
fifo_found_newdata       0       0     0       0       0      0           0
fifo_active_fifonodes  264     245   264     212     264    233         818
Time MRP (MM:SS)             14:41          02:34         14:40       14:41
Time CRP (MM:SS)             08:23          00:58         08:17       00:46   
----------------------------------------------------------------------------

Important : The results become much better after assigning the 0 value to the 
	    fifo_do_adaptive variable.

Important : The results are even better when we change from named pipes to
	    sockets. Doing that MRP takes 1 minute and 55 seconds to run and 
	    CRP spends 44 seconds.


			Best regards,

				Leo

8203.9MANMAN/X also uses named pipes by default!TROOA::HANDYL. Handy, MCSE DTN 626-3210Thu Feb 20 1997 15:3216
    We are currently running a MANMAN/X v3.0f benchmark, and our results are 
    VERY similar to .0 .  Note that MANMAN/X is now a CA product, and
    shares the same lineage as BAAN Triton.
    
    Yes, I have confirmed that we are using named pipes for mmx, as the
    default as described in the rpc_info init file.  Based on the results
    from .8, we are considering enabling sockets instead to improve elapsed
    time.  There are no cpu or I/O bottlenecks, but currently, under the
    named pipes configuration, the application appears to be the
    bottleneck, using only 10-20% of available cpu (4100, 2x400MHz cpu's).
    
    Any tuning recommendations?  According to CA, sockets is supported, but
    I don't believe we (Digital) have a reference site of MANMAN/X on
    Digital Unix to verify.  
    
    Lyndon
8203.10KITCHE::schottEric R. Schott USG Product ManagementTue Feb 25 1997 14:5118
Hi

 I suggest you first try:

   dbx -k /vmunix
      ass fifo_do_adaptive=0
      p fifo_stats

      run test

      p fifo_stats


If this improves your performance (using pipes) then please
file a qar or IPMT asap.

Note it helped the previous test quite a bit.