Next: Applications Characteristics that
Up:
Characteristics of HPC
Previous: Introduction
The consensus of the group was that there was a great need:
- for code development tools;
- for greater reliability of HPC systems;
- for code migration tools;
- to reduce system software inefficiencies;
- for looking at exciting and innovative applications areas, (to
help the HPC industry by stimulating new demands). This might
involve very data intensive applications (in contradistinction to
compute intensive ones) but also harder and more complex problems,
irregular data structures and less obviously load balance-able
problems;
- to generally increase market confidence in HPC as technology
that is becoming mainstream and that is now robust enough for real
commercial and industrial applications.
These technical and economic needs are strongly correlated. The group
believes that the technical deficiencies and general unreliability of
HPC platforms are the key to why industry and commerce are still only
gradually up-taking HPC technology in mainstream and core business
activities. However, the technical deficiencies are largely due to
insufficient commercial funding and industry-driven backing. It is
not clear how this loop can be broken as HPC alone is probably too
small an area to viably support commercial strength software.
Government intervention in the form of programmes like EUROPORT or the
UK's Parallel Applications Programme have been suggested as
possibilities.
The group identified a number of general observations about HPC
technology and Applications:
- It was felt generally that the HPC industry is perceived to be
in a state of financial instability and that this is highly detrimental
to long term planning both for developers and end-users.
- HPC as a field is technology-led rather than
applications-driven at present. This is true internationally and is a
serious concern for its long term future.
- The group felt that in the current maturing phase for HPC,
vendors and developers should be starting with applications and not
with the technology per se.
- We noted changes in the balance between usage of: workstation;
mid range platforms such as departmental-sized compute servers or
mainframes; and the very large supercomputer range of systems. These
three sectors overlap to a significant extent and compete with each
other and price/performance dictates the balance in user demand
between sectors. We note a contrast in trends between USA and Europe
in 5.
- New code development efforts (eg new companies) are more likely to
adopt HPC than companies who already have (large) legacy codes.
- We noted we had an insufficient number of commercial end-users in the
group. This reflects the limited number of active participants in the
HPC field who are from industry and commerce. How do we encourage
more industrial and commercial participation?
- A viable HPC vendor and software tool industry needs many viable HPC
applications. There is a need to broaden the application base outside
science and technology (since S&T is too small to sustain a viable HPC
industry alone). Perhaps education for industrial decision makers in HPC
exploitation needs greater attention and funding? Might funding be
made available for HPC academics to participate in end-user community
symposia such as in geophysics, chemistry, aerospace meetings?
This would carry the message to the end-users.
- Can the HPCC development community set standards without full
input from end-user communities and have the ``tail wagging the dog''?
HPCC is too important a technology to be controlled solely by its developers.
- as a group of international composition we note the dilemma of
duplicated efforts between Europe and the USA as well as increasing
input from the far east. To what extent is it healthy for HPCC
developments to be reinvented internationally? Should national
competitiveness be left to the end-user fields? There exists a clear trade
off: competitiveness of the US computer-making industry vs. US
computer-user industry. At present the latter may be suffering at
expense of the former.
Next: Applications Characteristics that
Up:
Characteristics of HPC
Previous: Introduction
Geoffrey Fox, Northeast Parallel Architectures Center at Syracuse University, gcf@npac.syr.edu