Conference abbott::postmaster

        Is this the Navy site formally managed by Paul D?  If so, have
        they ever corrected all the network problems they had with DDS? 
        That was their MAJOR problem.  They had DDS believing it was
        running on Alphas.
        
        In short they should not expect upgrading VAXen will fix DEPS
        because when it takes minutes for Message Router to process a DDS 
        update.
        
        Dave Zaniewski

    Yes, this is Paul D's old system. Yes, the network problems have been 
    resolved.
    
    

RE: <<< Note 1274.0 by VMSNET::M_NEVINS >>>

>Our customer, US Navy, would like to find out if upgrading to a faster CPU
>with more memory, might improve the performance when the DEPS1 process runs?

I would not be surprised if a faster CPU and more memory would speed up the
directory sync processing on the DEPS side. However, I don't know a lot about
DDS to know if it's likely to be disk bound in any case.
						-- Tom

Hello,

    You did not mention what the pertinent DEPS account and SYSGEN parameters
were set to on both the "fast" and "slow" systems: WSEXTENT, WSMAX, and
PGFLQUOTA/VIRTUALPAGECNT should be raised as high as possible.  Unfortunately,
we don't have performance data in the form of "If you have 50,000 DDS
entries, set these parameters to be this much."  But here are some guidelines 
that have worked for other customers:

   9000 DDS entries:
     WSEXTENT/WSMAX - 20000 to 30000 minimum
     PGFLQUOTA - 150000, or equal to the SYSGEN parameter VIRTUALPAGECNT
            *** The largest amount of memory that a process can
		map is limited by VIRTUALPAGECNT. ***

   22,000 DDS entries:
     WSEXTENT/WSMAX  40,000
     PGFLQUOTA/VIRTUALPAGECNT 300,000  

   27,000 DDS entries:
     WSEXTENT/WSMAX  35000 
     PGFLQUOTA/VIRTUALPAGECNT 66000


12 hours seems a very long time to process 50K DDS entries.
I suggest the following...

1) Make sure the above parameters are set as high as possible.

2) You mentioned that the "fast system" is a DDS World Search node.
Is the slow system a DDS World Search node?   If not, it may be a slow
link between that machine and the DDS World Search node that is taking
so much time, and I would suggest making the slow system a DDS World
Search node.

3) I recommend having a person with VMS performance expertise analyze the
slow system while the DEPS DDS extract is occurring to determine the bottleneck.
The slow process could be CPU bound, I/O bound and/or memory bound.  
After identifying the bottleneck, an educated decision about what hardware to
upgrade can be made.


Hope this helps,

       Mary