[for ALL] Valentina Server 1.x // And they say OpenBase is good?

[Ruslan Zasukhin]

on 2/10/03 9:51 PM, Frank Bitterlich at bitterlich at gsco.de wrote:

    PLEASE READ up to the end of this letter.

> LOL... Ruslan, I can see your point.
> 
> I had to make the choice for our business DB backend recently, and
> OpenBase was one of the two products I was testing. I got similar
> results to yours, but I'm not sure if that really counts, since I didn't
> look to deep into the possibilities on how to optimize it.
> 
> In the end I chose the other product (Frontbase), but mostly because it
> adheres very stict to the SQL92 standard. But the performance results I
> got were almost the same: Not even the same league as Valentina.
> 
> There are two things I wrote that off to: First, it is a server DBMS,
> and I expect it to have a more linear performance (read: not getting
> slower) with multiple, concurrent accesses. I have no way to know how
> Valentina Server will behave inthis respect,, but as it was originally
> developer as as single user DB, I think you might run into problems in
> this respect. But I'm open for surprises :)
> 
> The second factor is the cache and index optimisations. In my 500 000
> record test database, I got speed differences between 20 seconds and 0.1
> seconds for the same query, depending on cache situation and -settings.
> I'm confident that when I go into production with this database, I will
> be able to tune it (with the help of the Frontbase guys, whose support
> is almost as good as yours :) to much greater speed.
> 
> But still Valentina continues to amaze me. Today I had a corrupted ValDB
> (well, not really corrupted, but garbage data inside) and had to do an
> XML dump/edit/load to fix it. When I did the dump of the 40 000 records,
> I was waiting, and waiting, and waiting for the export to begin, not
> realizing that it all was already done in a fraction of a second. :) I
> think even BBEdit took longer to save the text file after I had edited it...
> 
> Maybe you want to contact the Openbase guys and ask them if you could
> tune your test DB some way...

:-) and pay them tons of bucks for "support"?

1) I believe this was VERY simple test. Just one table with all numeric
fields. What here can be tuned. I do not see in prefs option for cache
size...In any case it must go best from the start.

2) Frank, you probably have not note that right now we here totally rewrite
Valentina kernel from scratch.

In fact we do this not to fit multi-user model, as somebody can think.
No, I do this just to get better C++ design. I love perfect things...
When I did start I have not use some C++ features because in that time they
was very new and not all compilers have support them.

And I many times have repeat here, "single user DB" and "multi-user db" do
not differ significantly...Well, may be only Oracle has crazy complex file
format...and I am not sure they are right...

3) Frank, note, that increasing of cache as you describe, IS NOT
optimization. This is just usage of more fast device.
But be careful. This means that as only your db size become bigger of size
of RAM, you again immediately jump to 20 seconds speed...Actually even
worse, because again, as shown on my benchmark pages, speed of regular RDBMS
DO NOT grow linear to the number of records. IT is much worse, when size of
db is much more of RAM.

4) When you test db on OS X or Windows, virtual system can give surprises.
You can even quite app, start it again, but next query will be much faster,
because in RAM still present pages of db file. In fact OS work here as
secondary cache...
    So it is really hard to test it on cool system.
So you could see this effect, and think that you see speed up thanks to
cache size. You see?


***********************************************************************
5) And to warm up you :-)

We here already have Valentina Server working on Windows for now.
As backend Igor use 1.x kernel (!)
In other words we already have done:
    -- threading model
    -- sockets,
    -- network protocol

We can send all SQL commands of Valentina 1.x
And read/write data...

We have here crazy idea...May be we may ship Valentina Server 1.x ?
It will be based on current kernel.
No transactions. No new SQL parser... No Users/Group.
But this will be multi-user, true Client-Server access to Valentina
database!

What you think?  Should we do this?

***********************************************************************

-- 
Best regards,
Ruslan Zasukhin      [ I feel the need...the need for speed ]
-------------------------------------------------------------
e-mail: ruslan at paradigmasoft.com
web: http://www.paradigmasoft.com

To subscribe to the Valentina mail list go to:
http://listserv.macserve.net/mailman/listinfo/valentina
-------------------------------------------------------------