Conference ulysse::rdb_vms_competition

79.0. "Sizing Vaxen 4 Ingres!" by CGOS01::GBARNABE (SWS/Canada, way up North...) Tue Mar 01 1988 18:34

I am soliciting comments on the following recommendation:

A customer of ours has the following configuration:

+------+  +------+  +------+  +------+  +------+  
|750   |  |uVaxII|  |uVaxII|  |uVaxII|  |uVaxII|  
+--++--+  +------+  +------+  +------+  +------+  
   ||        |          |         |        |
   |+--------+----------+---------+--------+--> Ethernet
   |
   +---------+----------+------> Async DECnet
   |         |          |         
+--+---+  +------+  +------+  
|uVaxII|  |uVaxII|  |uVaxII|               <-- remote systems
+------+  +------+  +------+  		

- each uVAX is running a separate independent Ingres database
- users come in on terminal lines (not DECservers)
- the 750 does not have a database on it, but it is used for DECnet routing
- systems either have a couple of RD31's or single RA81, and about 9MB each
- there are approx 65 users spread over the 7 uVAXs
- overall average CPU utilization is about 40%, with 1 disk i/o per sec per user
- however they do experience daily CPU peaks of 90% for 1 hour duration
- the Ingres system consists of 2 part, for each user:
	- the front end: forms, query language, etc 
	- the back end: performs all the disk I/O
	- 1/3 of resources required for front end; 2/3 for back end
	- right now, front/back end are running on same uVAX
- customer anticipates approx 30-50% growth over a 2-3 year period

PROBLEM------------------------------------------------------------------

Customer would like to consolidate all the information from all the 
separate databases in a large database.
Systems are also having some performance problems occasionally.
As usual, price and anticipated performance are issues.  

DISCUSSION--------------------------------------------------------------

The customer is considering 2 alternatives.  Both scenarios deal with
centralizing the systems into one computer room.

--------------
Alternative 1:
--------------

Trade in most of the uVAXes on a few 3xxx systems, hooked in a LAVC.  One 
of the 3xxx systems would handle all the back end processes (about 65) and 
the other 3xxx systems would handle the front ends.

All the database would then be united on one uVAX.

The concerns I have is the perceived performance of an LAVC.  There could 
be considerable activity on the Ethernet.  As well, there could be 
considerable activity on the disk subsystem on the main 3xxx system.

--------------
Alternative 2:
--------------

Trade in most of the equipment on one BI based machine (eg 8550).

There is basically one problem here: price.  The customer is looking at 
Canadian $900k+ for a 8530.


SUMMARY----------------------------------------------------------------------

The customer is looking to deliver a recommendation shortly to higher
management.  A major concern is the price of any solution.  Unless we could
perform some collection of performance data (via SPM), it will be difficult
for us to recommend a solution. 

The concern over the LAVC solution is the possibility of acquiring yet 
another high risk configuration.   Could a 3xxx handle the back end 
processes?  Could the Ethernet handle the Ingres requests between the back 
end and the front ends?  (Ingres uses DECnet sessions to converse between 
front and back end.)

Ideally they prefer the BI solution, as it satisfies their primary goal.
It would also allow them to enter the world of BIG clusters in the long run. 

The question then becomes, what size of VAX?  It appears that minimally a 
8550 would be required, as the application is already distributed on 7 
VUPs (7 times 1 Vax Unit of Processing - a 780).  

This situation seems to be ideally suited to the new Calypso system.  The 
Calypso is a CVAX based BI multi-cpu system.  It would afford them the 
luxury of starting off with a 5-7 VUP system, and expand it upwards, over 
the next few years.

Unfortunately, I have heard that non disclosures for Calypso are not 
available...

(The customer would be willing to hold off till the fall for new systems.  They
do have a 750 that they could put online to hold them over.)

Any ideas anyone?

-- thanx
   Guy

    Guy, you will probably want to go with the larger central machine.
    I have seen Ingres running on a clients system and it is a very
    big memory pig!  An 8650 with roughly 80MB of memory had about 1500
    pages of memory left on the free list for around 50 users (working
    sets for the three processes required per person were roughly 400
    to 2048 pages).  Also it tended to make the CPU pant, so and 8530
    would be the minimum.  The client mentioned is considering an 8800
    upgrade to run Ingres.  Some system tuning was indicated and improved
    matters slightly.
    
    To re-state 79.1 - Why not RDB?
    
    								Neil