X-MimeOLE: Produced By Microsoft Exchange V6.5
Received: by onstor-exch02.onstor.net 
	id <01C78B92.552947DA@onstor-exch02.onstor.net>; Mon, 30 Apr 2007 18:45:01 -0700
MIME-Version: 1.0
Content-Type: text/plain;
	charset="us-ascii"
Content-Transfer-Encoding: quoted-printable
Content-class: urn:content-classes:message
Subject: Delorean Upgrade Discussion
Date: Mon, 30 Apr 2007 18:45:00 -0700
Message-ID: <BB375AF679D4A34E9CA8DFA650E2B04E03797129@onstor-exch02.onstor.net>
X-MS-Has-Attach: 
X-MS-TNEF-Correlator: 
Thread-Topic: Delorean Upgrade Discussion
Thread-Index: AceLklTnRQcF7PKvQjOlCRWAOQAuvA==
From: "Jay Michlin" <jay.michlin@onstor.com>
To: "Tim Gardner" <tim.gardner@onstor.com>,
	"Larry Scheer" <larry.scheer@onstor.com>,
	"Andy Sharp" <andy.sharp@onstor.com>
Cc: "Brian DeForest" <brian.deforest@onstor.com>,
	"Jerry Lopatin" <jerry.lopatin@onstor.com>

Tim, Larry, Andy,

Here is a report on the meeting Jerry held today to discuss Delorean
upgrades. This is a hard topic as a business matter just as it has been
for us as a development matter. So it was a difficult meeting too, but I
think the thinking was good and the outcome reasonable. But we will have
some more development work to do, which I'll explain below.

For the record, attendees were Jerry, Andy (LeFebvre), Caeli, Narayan,
Sandrine, Ed, Paul, Joe, Brian and me.

We emerged with three current options, and a request to explore a bit
more.

Option 1: Upgrade all from flash. Joe reports the cost of a CF is $25
($32 for the 1G part). There will be roughly an additional $200 per
customer cost for our processing and roughly $14 for shipping to each
site. So, for example, shipping 512M flashes to a customer with 4
Bobcats, two at each of two sites would cost 4 * $25 for the parts +
$200 for the customer plus 2 * $14 for the two sites, for a total of
$328.

Extend that to all 500+ of the Bobcats in the field at 250 customers and
the cost could come to $50,000.

Worse than the cost are the logistics. There would quickly be upgrade
and patch releases (3.0.1, 3.0.2 etc.), and we would need a way to
decide which we ship on flash and which we ship via download. We would
need to keep track of which Bobcats at which sites for which customers
were in each possible state. Granted that in the long term steady state,
everything would be smooth, but there would be a lot of churn, a lot of
management and a lot of cost and opportunity for trouble in the
meantime.

Option 2: Use the current two step upgrade process programmed in
Delorean. Jerry points out that he genuinely understands that this is
not a huge amount worse than the previous one step process which is
itself unpleasant. But he points out that even that process stretches
what we can reasonably ask of a customer. I'll add editorially that this
is an issue that someone should have raised a long time ago, and it's so
unpleasant now only because it is coming at the tail end of a
development project where it was not raised in the beginning.

Option 3: Write the standalone one step upgrade program. The attendees
understood that this will probably take 4 weeks of development time and
2 weeks of testing, that Delorean would be delayed waiting for it, that
it would add to instability, and that it would delay Cougar probably by
4 weeks. I'll add editorially that while the group understands these
things now, it will be easy to forget them later, so we will want to
make sure any decision involving this option is properly made.

All the options are unpleasant, so the reasonable decision is to ship
initially with flash (option 1), but at the same time begin work on the
one step upgrade program, option 3. While I don't personally love this
decision, I can't come up with a better one, and I respect the difficult
tradeoff it represents.

But there is more. If we are going to develop a one step upgrade
program, we ought to consider one that is quicker than what we have. Our
proposal to date has been for a one step program that would run perhaps
2-1/2 hours. Its virtue would be that it could run unattended. But could
we develop something that would run faster, say in 45 minutes? Or 5
minutes? As best I know, we have not explored that, and we are now being
asked to do an initial assessment and report it at the Delorean meeting
on Thursday.

I know that doing such work for our existing BSD system, just as we
expect to move to Linux, seems like a waste of effort. But our
responsibility is to keep the faith with 250+ customers, and if this is
what they need, we ought to look into it. I also know that there may be
limits to what we can do, in terms of software, stability and
development time. May I ask you to please think through these things so
we can present our thinking on Thursday.

In addition, Paul thought that a one step upgrade ought to be scriptable
and asked Sandrine to look into that. Since we looked into it and
decided it would not work, I'm not too optimistic. But it does not hurt
for someone else to look, so please be prepared for Sandrine (or maybe
Vikas) to come seeking your advice on such a project.

SUMMARY
1. We ship initial Delorean on flash.
2. We need to get to a one step install as soon as we can afterward.
3. It is an acceptable business decision to delay Cougar and other work
to achieve this.
4. If we're going to develop a one step insatller anyway, we need to
explore doing a good one, namely one that runs much faster than our
current proposal (but does not sacrifice stability).
5. We'll discuss this all at the Delorean meeting Thursday.

Apologies for the long message, and I hope you found it useful to come
fully aboard the discussion. I'll plan to make time tomorrow (Tuesday)
to discuss this personally with you all.

jay


