HI Joachim
Thank you for your quick answer.
Actually your answer on the first question makes a second question
irrelevant.
Actually what I tried to ask in the second question is following:
Let's say we have 10 tables, randata_1, rundata_2..... rundata_n and I
would like to compare the data in all tables. I can't do it in SQL. I
need first to unite all tables:
"select * from rundata-1 union all select * from rundata_2 ..... etc)
And if I don't need to unite all of them but only tables that were
created in past two days this will become more complicated. First I need
to choose what tables I need to unite from rundata_once and then
generate SQL script which will unite only these tables.
As I said your update makes this question irrelevant. New database
structure (which is right SQL structure for such database) solves all
these problems.
I do want to access database directly. What we are trying to do is to
export all data to Oracle so we will be able to use our reporting tools.
The main problem was database structure. Now when all run results can be
stored in one table this can be done easy
Thank You
Evgeny
-----Original Message-----
From: Joachim Worringen [mailto:
joachim@...]
Sent: Thursday, April 06, 2006 12:13 PM
To: Evgeny Kagan
Cc:
users@...
Subject: Re: Testing Perfbase
Evgeny Kagan wrote:
> Joachim Hello
>
> My name is Evgeny and I am Mellanox IT developer. Lately I have
received
> a small project, to create a web based indicators over perfbase
database.
>
> Can I ask you for questions?
Hi Evgeny,
sure - but better use the mailing list
users@... for
communication in order to make it useful for others, too. I cc'ed this
reply to
the list. In order to be able to use this mailinglist, you need to be
subscribed
which I have just done for you (using the address
evgeny@...).
> 1) Database structure.
>
> a. Currently each run results stored in new table. Is this
> configurable? For example instead tables
> (rundata_1,randata_2....rundata_n) can all results will be stored in
one
> table with proper index (1,2....n)?
This has already changed in the svn-version of perfbase: version 5 of
the
database format incurs exactly this change. The data (with
occurrence="multiple") of all runs is stored in the table "rundata" with
an
index denoting the run. Existing experiments with database versions < 5
can be
upgraded online using "perfbase check --update".
> b. If this is not configurable how do you unite the data to
> provide reports that comparing different runs? Do you make some unites
> in python and not sql
I do not fully understand your question, maybe you can explain your
specific
problem? You can differentiate data from different runs using the
content of
parameter values. I.e., you have a parameter value that describes the
version of
your IB software stack, then you can tell perfbase to show the related
data of
version 1.2 and 1.3. A simple example for this can be found in
examples/mpptest,
where two different MPI implementations are compared.
There also are other ways to differentiate runs in perfbase queries,
like
directly via the run index, via the time stamp, or the synopsis.
Generally, it seems that you want to access the database directly, not
using a
perfbase tool? I don't recommend this. Everything should be done through
the
perfbase commands like "query", "find", "ls", "info" and so on. If you
think
this is not feasible, let us know why.
Joachim
--
Joachim Worringen - NEC C&C research lab St.Augustin
fon +49-2241-9252.20 - fax .99 -
http://www.ccrl-nece.de---------------------------------------------------------------------
To unsubscribe, e-mail:
users-unsubscribe@...
For additional commands, e-mail:
users-help@...
