Jim Starkey wrote:

> = m. Th = wrote:
>
>> Hi,
>>
>> This is a draft about Firebird clustering, feel free to read and
>> comment.
>>
>
> I look forward to reading this. An initial comment, however, is that it
> would be useful to explain the problem that you want clusters to solve.
> It is probably obvious to you, but it isn't to everyone.

Yes, you're right. :)

> Among the
> possibilities are:
>
> 1. High availability (my guess is that this is your goal)
>

Yes, you're right :))

> 2. High throughput
>

In a way, you're right :))
The main goal is indeed to have a HA clustering without loosing any
byte. Now, we are in a rather good situation _if_ the Firebird team will
integrate in HEAD the remote shadows feature which behaved above of (my)
expectations - but this feature looses the last uncommitted work and
requires application restart). But if the clients uses different nodes
as SELECT sources we can have also a performance improvement.

> 3. I bought one that I need to justify to my boss
>

No, you are wrong ( :))) )

> 4. I need to justify a cluster to my become he'll let me buy one
>

Geee.... BTW, we are talking about open source isn't it?...

> 5. I really need a PhD in computer science
>

Oh, no I left behind this plane of existence many years ago ( :) )...
Thanks anyway for your kind thoughts...

> 6. I need something sexy for a grant proposal
>
>

My Borland/CodeGear Developer Studio 2006 is a very nice product which
starts very nice but when I close it a message appears: "Catastrophic
failure". Anyway I do my work with it :). You gain 1,5 points from 6.
Not bad, not bad...

> Clustering in itself isn't a goal.

Indeed. IMHO, Is a way to solve a problem. Not a Holy Graal. Holy Graal
doesn't exist here on earth.

> DEC got into clusters for two
> reasons: Tandem scared them out of their minds, and CPU engineering was
> unable to build a fast VAX. Neither of there justifications had
> "legs". MySQL's cluster product, on the other hand, was designed
> specifically to run inside of a cell phone switch.
>
>

Also, I'm scared about clusters. I don't like complex configurations
with distributed lock managers with many configuration ways to do it. I
started another HA project and I simulate some things in Delphi in our
apps and I saw that this can be done also in Firebird. I don't want a
general cluster capable to do anything on paper and (almost) nothing in
reality. That's why I try to keep the things simple. I agree with you in
the way of thinking.

> Making the requirement an explicit part of a proposal tends to keep a
> project on track towards a stated goal.
>
>

High availability. Hot-swapping the data servers. (Near to) 0 points of
failure. Transparency and ease of configuration. If, as an aside, we
achieve a throughput increase, so far so good.

> As the first person to implement a cluster based database design
> (Rdb/ELN), I've rather lost faith in cluster technology. Like invading
> other countries, it looks good on paper, but the details and
> implementation ramifications makes achieving the stated goal very difficult.
>
>
>

Then perhaps you can help now? :) In any circumstance, the experience of
the elders is invaluable

>

TIA,

m. th.