[asac]Re: Draft of ASAC Report: section on RSC

[K. Tatematsu]

Dear Pierre, and ASAC members,

At 17:46 02/04/02 +0200, Pierre.Cox at ias.u-psud.fr wrote:
>We note that the role in supporting archival research overlaps to some degree
>with the role of Virtual Observatories and the interaction between the RSC
>and these VOs needs to be defined. Whether each RSC has a physical copy of the
>archive or merely provides a link is a matter of implementation. Various
>communities have considered  other functions to be important (e.g., financial
>support for the US community, software development for the Canadian community),
>but the core functions listed above have been agreed to be the common 
>denominator.

I have a comment on the sentence "Whether each RSC has a physical copy of the
archive or merely provides a link is a matter of implementation."
# Although I said the same in the face-to-face meeting, I am afraid that 
most of you
# did not understand my comment.

Let's think about doing archival astronomy by using VO.
Ten years later, we will have local ALMA archive (physical or link) and 
Subaru archive in Tokyo.
If you believe in Moore's law in computing, the network speed will be of 
order 100-1000
times faster than the present day in 10 years.
So, local network speed (within NAOJ and directly connected universities)
would be of order Tbps (now 1-10Gbps), while global network
between Tokyo and Chile (or USA or Europe) would be of order Gbps (now 
1-10 Mbps).

Some astronomer may try to do VO astronomy (without using pre-existing simple
catalogues) to get brand-new result by cross-correlation.
We assume that he/she likes to cross-correlate the ALMA and Subaru database.
You know that yearly ALMA data amount can be
of order 1 PB, and the archive is very large in size.

If the ALMA archive is physical in Japan, cross-correlation on network is 
very fast on Tbps
network. On the other hand, if the ALMA-J archive is just a link to Chile main
archive, cross-correlation is 1000 times slower, because it uses Gbps line.

Even for a USA archive researcher has benefit from high-speed link between
the ALMA-J physical archive and Subaru archive.  Cross-correlation
will need high-speed network between archives, while the result
can be sent to USA even on slower Gbps line because the data size of
cross-correlation is much smaller.

Here, I assume that cross-correlation is not CPU-power bound
(e.g. position of two data archive is well determined, and image matching 
is not necessary
...)

I feel that whether the local archive is physical or just a link is not an 
implementation
issue but more serious if we mind how quick we can get scientific result.
Or, you can wait for 1000 times longer time...

If I miss the point, please let me know it.

Cheers,
Ken