Re: [ROOT] Re: TTree modification

From: Anton Fokin (anton.fokin@smartquant.com)
Date: Tue Feb 27 2001 - 10:52:23 MET


Hi Pasha,

right, in our experiments we also never overwrite primary dataset which
contains everything, including lots of garbage. We did it with tapes/PAW and
we do it now with HDD/ROOT. That's OK. Let's assume that an "average
customer" would also keep primary data in a safe place. On the other hand it
is hard to understand why secondary dataset should be overwritten completely
if I want to see how results change after changing one field in one record.
Also, it is a bit hard to argue "why any SQL database has INSERT and DELETE
commands". I personally prefer to have both possibilities (overwrite or not)
even if I would not use overwrite in 99.9% of my data manipulation. On the
other hand I do not request to implement something which will affect ROOT IO
performance. I asked this question just because Rene mentioned several times
that overwriting is quite easy and is not introduced because of political
reasons. I am also a bit curious why ROOT has only one storage class, but
that's another issue.

Regards,
Anton

http://www.smartquant.com


----- Original Message -----
From: Pasha Murat (630)840-8237@169G <murat@ncdf41.fnal.gov>
To: Anton Fokin <anton.fokin@smartquant.com>
Cc: roottalk <roottalk@pcroot.cern.ch>; Eddy Offermann <eddy@rentec.com>
Sent: Tuesday, February 27, 2001 2:46 AM
Subject: Re: [ROOT] Re: TTree modification


> Anton Fokin wrote:
> ... snip...
> >                                                    But it is quite hard
to
> > explain "one write many reads" issue to an average customer of the
> > framework, which suppose to be general enough to serve people doing
> > different things (including university researchers in finance for
example).
>
> hi Anton - didn't you try to explain to an "average customer" something
along
> the following lines:
>
> the data which are coming in in real time and can't be
> reproduced later SHOULD NEVER be overwritten/corrected. This is what is
called
> primary datasets in HEP. THis is why "one write many reads" is a very
natural
> concept for the experimental physics. We also have a notion of a secondary
dataset
> which is a derivative (selection/result of reprocessing) from the primary
dataset.
> The secondary datasets, unlike the primary ones, can be recreated as many
times as
> it is needed, so correcting the "raw" data never becomes an issue.
>
> best, Pasha
>
>



This archive was generated by hypermail 2b29 : Tue Jan 01 2002 - 17:50:38 MET