Conference mvblab::alphaserver_4100

435.0. "The Cluster Company?" by NQOS01::16.137.80.18::Workbench () Mon Jan 27 1997 11:10

This is a question for those who designed the hardware architecture for a 
4x00 cluster.  I welcome all corrections if my recent experiences did not 
give me an accurate picture.

We are supposed to be "the cluster company".  I recently had to quote a 4x00 
cluster against Sun.  This was going to be an ASE cluster.  The customer 
wants a rack, so I go thru the pianstaking process of quoting all the parts 
(you know, this "Y" cable and that "SCSI" cable, etc.).  After all this, I 
realize that brilliantly, the engineers who designed our ASE cluster for 
"high availablility" and "Continuous Computing" (one of our nine target 
market segments), designed in a single point of failure in that the bottom of 
the rack only has a SINGLE 208v power supply.  So I can configure n+1 power 
to every component in the cluster but when I put the components into the rack 
and plug them into the power suply, my customer's highly available solution 
is dead if the SINGLE power supply croaks.  I am told that a second power 
supply will not fit (I maxed out the storage shelves, but there is not even a 
seperate part number for the power supply that either I or DECsale could 
find).

The Sun quote came back the same day, with much less complication (a standard 
"cluster" quote with the variables highlighted - CPU speed, memory, etc) and 
no single point of failure.

I am told I have to go to CSS to get a cluster in a rack that has no single 
point of failure.  I am also told by CSS that the rack designs for the 4x00 
from the SBU did not figure in much of the experiences that the CSS group has 
with building rack mounted systems (like which way the fans blow, etc.).

If it seems like I'm a bit irritated, I am.  But I am probably wrong in some 
of my assumptions and conclusions since I admit to not knowing everything (a 
rare but much sought after quality...).

Can someone in the 4x00 product group explain any of this?

Bottom line:

1. When will I be able to quote a "cluster" without having to re-invent the 
wheel by quoting the detail components?

2. When will we live up to the "Digital is the cluster leader" badge that we 
wear by ensuring no single point of failure in a simple HA rack 
configuration?

    You may want to contact Joe Scala, a product manager in our group.  He
    is currently addressing this issue.
    
    Todd

First I would request that the person who entered this note reply with a real
employee name and not an anonymous entry. This notes file is NOT a soapbox for
people to get up and complain. Use it to convey problems for which you want a
response from engineering. That said....

The only 'cluster' program in the company has been developed by CSS. A part of
that program is going to be made more visible in a few weeks. Joe Scala is the
SBU product manager driving this effort. The goal is to make it easier to order
clusters. Some of the complexity of having to order MANY part numbers will be
reduced to having to order a few part numbers.

'Clusters in a single rack' are in reality a SMALL fraction of the business.
When the attributes of the current rack were determined by product management it
was determined to be a NON-GOAL to have to support dual AC power rails. They/we
were wrong! CSS had already defined the 'standard' rack to include dual AC
controllers. We are currently in the process of defining a blue rack that will
be the standard for CSS or SBU modular solutions. It will cost customers several
hundred dollars more but they will shortly get redundant AC front end
controllers!

Please send a copy of the SUN quote and your DIGITAL quote to me at:

Dennis Bak
129 Parker St
Maynard, Ma 01754
MS PK3-1/R11

I will make sure that Joe and the other product managers working the cluster
program issues see what your up against when you want to develop a quote for a
DIGITAL solution.

	Dennis Bak....Rawhide Systems Engineer