| T.R | Title | User | Personal
Name | Date | Lines |
|---|
| 321.1 | | KONING::KONING | Eesti vabaks! | Tue Aug 06 1991 17:11 | 4 |
| Well, as soon as the DEMFA is out, you can connect a few to a big ring
spread across town(s) -- with single mode fiber concentrators. Soon enough?
paul
|
| 321.2 | connectivity does not equal disaster tolerance | STAR::SALKEWICZ | It missed... therefore, I am | Thu Aug 08 1991 15:12 | 71 |
|
Well ,.. not quite.
Evaluation testing will be started shortly, but just having FDDI to XMI
connectivity does not mean you have disaster tolerance.
Right now, VMS supports only CI connections between 6000/9000 class
machines in a cluster. It is unsupported for example to connect two
6000's with Ethernet, and cluster them that way without any CI
connection. Yes it works,.. but it is unsupported. The reason it is
unsupported is because the horsepower of these big machines is high
enough to swamp Ethernets. It is not clear that it would swamp FDDI,
but its also not clear that it wouldn't. What also becomes a factor
is the amount of CPU horsepower required to keep a cluster operating
with reasonable performance at the user level. After all,.. our
customers aren't buying these high priced hunks of hardware just so
that all their CPU is consumed by clustering software,.. they expect
that some CPU should be left over to run their all-in-1 sessions too.
FDDI uses more CPU per request (as does Ethernet) because it is a
generic networking interface which requires the clustering software
to do lots of extra work (CPU work) compared to the work required to
driver a CI connection.
We are about to announce support for the DEMFA, which will include
support for FDDI as a cluster interconnect to the same level that we
currently support Ethernet as a cluster interconnect. What that means
is that it would still be unsupported to cluster two 6000's together
using FDDDI/Ethernet only and no CI.
Test plans have been set up to evaluate the possibility of relaxing
these restrictions for FDDI clusters. When the tsting is done, based
on the results, we may indeed declare support for FDDI as the sole
cluster interconect between large (6000/9000) machines. If/When that
support is declared, then the WAVC (Wide Area Vax Cluster) or MAVC
(Metropolitan Area Vax cluster) will be a reality,.. but still not
necessarily disaster tolerant.
It has already been determined that disaster tolerance will require
additional engineering work to be done. Its not what I would call
a major project, but a project nonetheless. The engineering and
testing will be done soon, and support for disaster tolerance
may become a reality at that time.
The long and short of it is:
1) We only have support to the "ethernet replacement" level with
V5.4-3 pending successful completion of the field test currently
in progress.
2) We may get support to the "CI replacement" level pending
evaluation that is not yet complete. This will give us WAVC
or MAVC.
3) Disaster tolerance is still pie in the sky. Some work has yet to
be done before that becomes reality.
So in short,. we will probably get the MAVC/WAVC type of product
announced fairly soon after DEMFA is announced,... but disaster
tolerance will be a bit longer after that.
So how much of our lunch can we keep IBM from eating by simply
having WAVC/MAVC,.. and how much will they still be able to steal
until we get disaster tolerance?
Another question is,.. does their offering really provide disaster
tolerance? If it isn't painfully obvious by this time,,... simply
having the wide area interconect does not say anything about the
ability of the distributed system to tolerate/survive disasters.
/Bill
|
| 321.3 | My twopennorth | LARVAE::HARVEY | Baldly going into the unknown... | Fri Aug 09 1991 05:52 | 28 |
|
My experience with customers and their interest in FDDI Clustering
is that the majority would 'simply' like to do Volume Shadowing over
a wider geography. This helps to provide basic Disaster Tolerance as
at least they can have their data safe and ready to hand. They can
make a decision as to what work takes precedence on their remaining
systems as/if/when a disaster occurs.
Bear in mind these customers are not Banks and Finance houses with
wall-to-wall Fault Tolerant systems and where minutes of lost
systems time can mean loadsamoney - but more average customers with
limited budgets to spend... They can't afford to have their systems
running at 50% load just in case half their clustered systems go
AWOL.
I note Bill's comments with interest ie. Digital being cautious with
this technology and proving how much is possible before positioning
it as Disaster Tolerant. However, are you aware of (VAXCluster)
engineering plans which intend to launch the technology precisely
into this space - with associated LARGE licence fees ?!
Should the plans for FDDI Clusters (as I currently understand them)
are implemented I should forget trying to sell to normal customers
as the costs of licences are likely to be prohibitive.
Regards
Rog
|
| 321.4 | flash | STAR::SALKEWICZ | It missed... therefore, I am | Fri Aug 09 1991 16:14 | 30 |
| Yes I am aware of the VAXcluster engineering effort. They are slightly
ahead of themselves in my humble opinion. I am in support of making
money for the corporation,, and providing disaster tolerance would
be a big win obvioulsy.
Since you asked,.. I just went and got myself an update. It seems
that we are closer to making an offering in this space than I
thought. But some things have changed also.
The major new piece of information that I have is that the plan is
to require human intervention after the disaster to do the recovery.
Obviously, this is much less complex than having VMS/clusters do the
recovery automagically, (which is the only true disaster tolerant
system IMHO) with all shadopw sets and transaction
databases intact. SO what they are trying to sell is a bunch of
procedures for bringing a cluster back up after a disaster. Doesn't
really seem like the kind of product thats worth megabucks to me.
In any event, as soon as VMS declares support for the DEMFA, they
will try to launch a field test of the disaster tolerance stuff,
whatever form it takes. The current schedule calls for an announcement
in the October timeframe.
I have also learned that we will be relaxing the "must have CI between
big nodes to be supported" guideline. Rather than hazard a guess as to
how relaxed we're going to get, I'll enter a copy of the SPD when it
becomes available. Then we can all try to interpret it together :-)
/Bill
|
| 321.5 | I think we are on the right track... | KYOA::KOCH | It never hurts to ask... | Sat Aug 10 1991 11:22 | 23 |
| Well, at the VAXcluster Symposium I attended the presentation for the
Multi-Datacenter VAXcluster Systems (aka WA FDDI VAXcluster, Disaster
Tolerant VAXclusters, etc.). I think I am in the minority right now,
but I support the licensing fees concept. We apologize for the price of
our technology now and if I am correct, we are the only company in the
world who is going to be able offer this capability.
We have always had the problem where we think we know all the problems
and all the answers. We were wrong. In the case of the this new
technology, we assure the fact that only certain classes of customer
can acquire this technology day one. It will make our support job
easier. I am working with a customer to possibly field test this
product and when we said it would be a $1M cost, he said "So, WHEN CAN
I GET DELIVERY???".
In IBM announcement, I BELIEVE that they know we have a competitive
advantage and want to blunt it. I want to get a few good customers,
train the field, then reduce the price a little. Get the next tier of
customers, reduce the price, get the next tier of customer, on so and
so on.
Remember, you can always reduce the price. However, in most cases,
raising a price is always a heck of a lot harder...
|
| 321.6 | Learnings... | LARVAE::HARVEY | Baldly going into the unknown... | Mon Aug 12 1991 08:32 | 28 |
|
Re: .5
Fine, I appreciate the arguments (as per the Product Management
Business Plan etc.). However, I would ask that we all learn from
this experience. We seem to repeat such problems with regular
monotony !
I for one have been talking to customers (under PID) about FDDI
Clusters for about 6 months or so. In the PID presentation there is
NO warning/indication that this technology will cost the customer
(dearly !) with special licences etc. My understandings of this
scenario is that the decision(s) to position FDDI Clusters at the
high end was made quite recently by Product Management.
The result ? Raised customer expectations - despite the front page
disclaimer of the PIDs. Cluster users see the XMI adapter as opening
the door to FDDI Clustering (..."its just like LAVCs with more
bandwidth"...)
As I said earlier (.3 ?) I can forget trying to continue selling
FDDI to most of my customer contacts/accounts as their main reason
for purchase was to work a split site Cluster with Disk Shadowing.
I just wanted to flag this issue so that others don't fall foul of
the same problems.
Rog
|
| 321.7 | T3 addresses other markets | DELNI::GOLDSTEIN | Networks designed while-u-wait | Tue Aug 20 1991 18:00 | 15 |
| FDDI and T1 are not direct substitutes for one another!
IBM's offering T3 cannot be blunted by our FDDI. The former is a
tariffed telephone company service, the latter requires dedicated
glass. We have it easy in Littleton, where we can hang anything we
want on the poles if we get Curt's permission. (He's the electric
company.) Try and get a private FDDI cable around a typical city or
suburb, with many roads to cross, a recalcitrant utility, and already
crowded poles! If it were easier to hang your own wires all the time,
"the telephone company" wouldn't be so rich powerful.
FDDI's primarily a campus solution. A few lucky customers can go
beyond. T3 goes anywhere. For a price. FDDI's a lot cleaner and
cheaper when you don't have rights-of-way to worry about. T3 isn't
cheap, but when you need it, you can get it.
|