Hardware Reference
In-Depth Information
finds vast computing resources in the simulation centers of automotive OEMs
and their suppliers alike. Industrial customers have invested and continue to
invest in hardware, and expect software vendors to make the best use of it so
as to improve the Return on Investment (ROI) and reduce the Total Cost of
Ownership (TOC). For example, as microprocessor manufacturers are not pro-
ducing faster CPUs anymore but rather ones with multiple cores, customers
expect software to use multiple threads eciently.
20.3 The I/O Challenge
These changes in the modus operandi of industrial customers are reflected
on a broad set of requirements on the data access, manipulation, and storage,
what could collectively be called simulation data management. Such require-
ments, imposed on software vendors, have in turn increased the non-functional
requirements to the persistence systems used:
1. Data should never be lost. The "last known good version" should always
be accessible. More often than not, the development of die set geometries
is laborious and very costly to repeat. Delays in design also cause delays
in downstream processes, only to increase the costs.
2. Data should be accessible from normal users in the form of files. These
files are often manipulated by customer processes that back up, com-
press, delete them, or other processes like Product Lifecycle Management
(PLM) software that treats files like product assets, providing audit-
ing and tracking, load-balancing systems like Platform LSF, Condor, or
others. Furthermore, user account restrictions apply to the creation and
manipulation of files and the directories in which they reside.
3. Allow access to files from multiple threads and multiple processes: ap-
propriate coordination so that there is no data loss or corruption and
so that the hardware is optimally exploited. Parallel processes, either
on the same machine or on a network, need to access data using differ-
ent patterns; simulation software, as a data producer, creates new data
structures in a serial manner for a single simulation; processing software,
as the main data consumer, needs to have access to simulation data in
a lateral manner, spanning multiple simulations.
4. Files should be accessed eciently on network drives using drive map-
ping or other means that allow shared storage areas to use similar paths
as local storage media. Accessing files on shared storage areas increases
complexity as processes from different platforms access network file sys-
tems in different manners.
