Impact of Technology Upgradation - A Case Study of RCC,ONGC,Vadodara

Document Sample
Impact of Technology Upgradation - A Case Study of RCC,ONGC,Vadodara Powered By Docstoc
					                        Impact of Technology Upgradation
                     - A Case Study of RCC,ONGC,Vadodara
             • P. K. Mittal • M.C. Kandpal • Ravmder Mohan, RCC, ONGC, Vadodara

         Seismic processing has taken extraordinarily long strides during the last decade. From us-
ing mainframe systems consuming large resources in terms of power, air-conditioning requirments,
floor space and thus operational costs, the geophysicist has now moved to much more compact and
user - friendly systems. In terms of processing paradigm, batch processing has given way to a mix
of batch and interactive processing with more and more reliance on interactive and online outputs
of hitherto time-consuming processes such as deconvolution and various filters.

The expectations of the interpreter from the processing geophysicist have increased many-fold and,
aware of the fact that the processor has-immense compute power at his disposal, the interpreter
expects results in terms of days and hours rather than in months and weeks, as was the case in the
past. Technology enhancements have had tremendous impact on the very fabric and mind-set of
processing-interpreting-geophysicist combine. In this paper, we try to present a case study of the
impact of technology enhancement at the processing centre of ONGC, Vadodara in terms of tech-
nology advancements and new capabilities along with the associated expectations of management
as well as clients, challenges and steps taken for Change Management.


Technology Upgradation

         The centre was earlier equipped with a
mainframe Russian system (EC-1061) till 1999.
It was mainly used for 2D and VSP processing
as it did not have 3D processing capabilities. Sub-
sequently, the centre has been upgraded with the
installation of a 6 node IBM SP - 3 system with
significantly higher computational power (Table-
1). Besides having much greater peak computa-
tional power, memory, disk capacity etc. the sys-
tem has 38, RS- 6000 workstations with its own
memory, internal disk (9.1GB) and colour dis-
play capabilities. Additionally, there are two Sun
Ultra-60 workstations for interpretive, model
based processing and 3D visualization. The out-
put devices are the state-of the art 3590 cartridges
(10GB) along with 3480E’s, 3490E’s and nine-
track tape drives of ensure compatibility with the         processing capability enhancements, which have
earlier system outputs, and with other process-            got considerable implications on processing work
ing centres of ONGC. The centre has been                   culture, aspirations of interpreters as well as
equipped with online thermal plotters and colour           management in terms of marked improvements
plotters to various sizes (Fig. 1). The entire con-        in processing turn around and quality.
figuration of 6 nodes, workstations, tape/cartridge
drives as well as plotting/printing devices are            Processing Capability Enhancements:
networked through fast Ethernet (10/100mb/s)
and four Gigabit switches. All these hardware                     The hardware and software upgradation
upgradations have resulted in very significant             has obviously added a host of new capabilities
                                                       1
                                                                           GEOHORIZONS Vol .6 No. 1
                                                                                January 2001
                     Fig 1. Seismic Data Processing system at RCC, ONGC, Vadodara

including 3D processing with pre-stack time              below a good, correctable marker.
and depth migration options. Model-based pro-
cessing, interactive trace editing across chan-          Effect of Upgradation on Quality of
nels, fine grained interactive velocity picking
with almost - instantaneous checking of the
                                                         Seismic Outputs:
suitability of velocity picks by applying NMO,
velocity field generation and multi-velocity                    The upgraded seismic processing soft-
function stack generation are some of the very           ware has a number of modules/ algorithms which,
useful tools now in the hands of the processing          when applied judiciously, enhance the quality of
geophysicist. The 3D visualization (in colour)           processing at various stages, leading to improved
are useful tools for quality control during ge-          processed outputs. Figs 2 to 5 show a few ex-
ometry definitions and ensure early detection            amples of the enhancement of data quality after
of any errors, thus obliviating the need to go           processing using new processing software. Fig.
back, which has a direct impact on reducing              2 shows a seismic line passing through
the processing turn around time.                         Limbodra- Halisa area connecting the produc

        Now that the enhanced capabilities are
known to the interpreters/clients, their expec-
tations have risen to new -levels. Not only do
they expect outputs much faster and of much
better quality, it is natural that they would like
to see more colour displays, closer velocity
picks and more ‘on-screen’ interaction with
processors to see more test panels before deci-
sions on parameters are taken. The zone of in-
terest is more focused and sometimes they
would be looking for events just 200-300 msecs                 Fig 2. Seismic line ‘A’ : Halisa-Sadra area

                                                     2
                                                                           GEOHORIZONS Vol .6 No. 1
                                                                                January 2001
ing Limbodra field to Halisa in the eastern mar-
gin of Cambay basin. Deeper reflections below
Deccan Trap, occurring near 1100-1900 msecs
in the eastern part of the line could be brought
out using new algorithms. Fig. 3 shows a NNE-
SSW trending line in the same area where pro-
cessing has improved the section to an extent so
as to bring out two episodes of deeper reflec-
tions below Deccan Trap which occurs at 400-
500 msecs towards NNE. The reflection regime
starts from 600-800 msecs in the NE part and
can be seen up to the SW part of the line. Exact
location of faults and undulations on trap top,
                                                               Fig 4 : Seismic Line ‘C’ : Rajpipala-Rajpardi area
hitherto not visible, can be seen. Fig. 4 is a seis-
mic line in the Rajpipala-Rajpardi area in the
eastern margin of Cambay Basin, which shows
seismic data to the north and south of Narmada
reverse fault with probable wrench faulting
caused by Son-Narmada lineament and orthogo-
nal Dharwad trend as clearly marked by differ-
ent seismic characters at the two ends of Narmada
reverse fault. The quality enhancement of the
section was possible due to use of sophisticated
algorithms and involved processing steps avail-
able with_Geovecteur processing software.
                                                               Fig 5 : Seismic line ‘D’ : Kerala-Konkan Offshore
       Fig. 5 is a part of seismic line in Kerala-
Konkan offshore area reprocessed at RCC,                   Parallelization Issues:
Vadodara in order to remove multiples, water
bottom reverberations and coherent noise, which                     The number of computations required to
had masked the signal and could not be removed             process a seismic survey continue to increase.
when the line was processed earlier. The repro-            This is both because of the increase in the amount
cessed section shows atext book example of                 of data recorded and to the use of complitationally
progradational features between 1400 to 1600               intensive processes. Availability of a parallel pro-
msecs. This example again shows the effective-             cessing system is a boon to the geophysicist due
ness of the processing software and upgraded               to the speed available for solving data process-
computing facility:                                        ing problems. Any number of CPU’s that are con-
                                                           nected to one another can potentially be used to
                                                           perform parallel processing. Some of the com-
                                                           mon ways of connecting CPU’s (or nodes) are :-

                                                           -     Symmetric Multiprocessor System (SMP)
                                                                 Shared memory and disk
                                                                 Single copy of OS shared by all processors
                                                                 on the Bus or Cross Bar. However, the
                                                                 scalability is limited by the effectiveness
                                                                 of memory bus
                                                           -      Loosely coupled processors. In the case,
                                                                 disk is the only shared resource.
      Fig 3 : Seismic line ‘B’: Halisa-Sadra area          -     Massively parallel processors (MPP) In this
                                                       3
                                                                               GEOHORIZONS Vol .6 No. 1
                                                                                    January 2001
      architecture, nothing is shared between                   mix of both parallel and sequential processsing
      nodes. Each node has its own memory, copy                 in the same job set up is possible. A second level
      of OS and disk. Nodes are linked to other                 of parallelization is possible by restructuring the
      nodes by a high speed interconnected                      algorithm itself in order to parallelize the appli-
      switch. High degree of scalability can be                 cation. This is done for highly compute- inten-
      achieved if the application is designed to                sive applications, such as 3D Pro Stack Time
      take advantage of this architecture.Parallel              Migration, 3D Pre Stack Depth Migration, 3D
      processing efficiency is strongly dependent               Post Stack Depth Migration and 3D Post Stack
      on node- to-node I/O band width, cache and                F-XY Time Migration.
      latency.
                                                                        The efficiency of the 6-node SP3 system
                                                                to perform parallel seismic processing was tried
                                                                using 3D Post Stack time migration. The
                                                                parallelized version of this software (GTMIP) is
                                                                available in Geovecteur Plus processing software.
                                                                This module performs one pass 3D time migra-
                                                                tion after stack, using the wave equation in the
                                                                space-frequency domain. All the six processors
                                                                were used. Fig. 6. depicts the utilization of the
                                                                nodes during GTMIP run. A linear scale down
                                                                of elapsed time with respect to number of nodes
                                                                was observed.




 Fig 6 : Utilisation of al 6 nodes during parallelization
                    migration job run

        Seismic data processing is a very good
candidate for parallelization. It is achievable by
making use of units like traces, gathers or fre-
quencies. A wide range of applications can be
parallelized without having to re-code each ap-
plication. Most commercial seismic processing
softwares have this features. In CGG’s process-
ing package Geovecteur, modules are available
for parallel flow splitting, which split a job into
different processes in order to parallelize the pro-                 Fig 7 : Node Utilization for 24 hours period
cessing depending on the defined data flow, A
                                                            4
                                                                                  GEOHORIZONS Vol .6 No. 1
                                                                                       January 2001
Authorization Issues:                                           With the availability of a 6-node parallel
                                                        machine with a peak computational power of 4.8
        In the new environment there are two            GFlops against the earlier 8 MFlops, the power
stages for checking authorization. One, by UNIX         of the machine at the disposal of the processing
and second, through Geovecteur(by Job Man-              geophysicists is 600 times that of the earlier ma-
ager). UNIX has three levels of authorizations          chine. Even in spite of the increased overheads
viz. Owner, Group and Other. Geovecteur (GV)            of the new GUI based system, the geophysicist
users work on ‘Projects’. Groups of project hav-        can now get outputs of processes such as F-D
ing some commonality are classified under a             Migration for a 20 Km line with 25 m shot inter-
single ‘Charge’ Charges, projects and authorized        val as well as group interval, in about 5 minutes
users are all defined to the job manager, other-        as compared to earlier 10 hours or so. Similar is
wise a user will not be permitted to open his           the case for all other processes such as Decon,
project by Geopad, the GV front end. In GV there        Filter, Stack, etc. With these speeds the turn
exists a ‘Proj’ directory, under which charge di-       around time for complete processing of a nor-
rectories are created. Under each charge direc-         mal 2D line (with reasonably noise- free chan-
tory there may be many project directories.             nels and well-defined geometry) is considerably
Charge and project directories are created by the       reduced now. A line can now be completely pro-
GV super-user who is the ‘Owner’. Within Unix,          cessed well within a week as against a time span
a group is created corresponding to each charge.        of about 3 months in the past.
All users who would be using projects under a
charge are designated as members of this UNIX           System Utilization:
group. A Geovecteur user will be able to use a
project provided he is authorized by the job man-                In the days of the mainframe, it was the
ager and has the necessary Unix permissions. The        CPU time that was the most expensive. When a
charge directory filters access through Unix to         computer system was provided, the management
project directories. RCC, Vadodara has imple-           expected that CPU should be heavily utilized (at
mented a scheme for naming charges and projects         least more than 70% on a 24 hour basis). The
with a view to ensure easy access to information        system was expected to give maximum process-
pertaining to data processed.                           ing throughput and job-mix was mostly batch-
                                                        oriented. However, with the advent of new tech-
        Charge directories are named on a 4-digit       nology involving much faster speeds (our centre
code having format GSLD. G and S represent              has 4.8 Gflops peak compute power) and soft-
the processing gourp and subgroup respectively.         ware being much more interactive, two changes
L denotes whether the data is land-data are rep-        have occurred. One, it is not possible to keep the
resented by 1, 2 and 3 respectively. The project        CPUs fully engaged under normal processing-
directories are named on a 7-digit code in the          mix (this means we are not talking about PSDM),
format IIIASYY where III represented the seis-          even with multiple 3D prospects being in pipe-
mic investigation number. A is the area code, S         line. The second, somewhat paradoxical, situa-
denotes the type of source (explosive, vibrator,        tion is that the response time of the workstation
etc.) used, and YY indicated the year of process-       where the processing geophysicist is working
ing. Two digits are provided for year because the       with many interactive applications should be very
processing in the new environment has com-              good. Thus, delay of even 2-3 seconds is notice-
menced in the new millennium. With the nam-             able and is ‘irritating’ to the geophysicist who
ing convention followed, it is possible to access       now finds the system too ‘slow’ for his require-
information based on the group who processed            ment. This means that though the system is
the data, when the data was processed, the type         grossly ‘under utilized’ during off hours, the sys-
and area of data, source used and any combina-          tem appears to be too slow during the working
tions of the above.                                     hours. In other words, while justification of more
                                                        computing power is possible in terms of the pro-
Turn Around Time:                                       cessing applications, we have to live with the fact

                                                    5
                                                                         GEOHORIZONS Vol .6 No. 1
                                                                              January 2001
that on a 24-hour, 7 day week basis, the system            bottleneck and it was considered necessary to
is going to remain grossly under utilized (See             increase it. The software engineers who were
fig. 7).                                                   studying this problem began examining as to why
                                                           Geopad could not be run on other nodes. The
Multi-Node Utilization:                                    problem was found to be due to TCL (Tool Com-
         In Geovecteur Plus environment Job                mand Language) not being installed on other
manager controls the distributions of seismic              nodes. TCL was installed on all the other nodes.
tasks over the network or among nodes of a multi-          After this, the users could run interactive appli-
node machine. The node on which the main com-              cations from any node. Each user group was as-
ponent of Job Manager is installed is called the           signed a node for interactive applications such
‘Logger’ node and in our center, it is node I. It is       as trace editing. The interactive response im-
recommended to run a maximum of 3 jobs si-                 proved tremendously as also the utilization of
multaneously on a node to get optimum response.            the nodes.
Since node I’ is a ‘Logger’ machine, all actors
on it are closed except the one for running demux          Conclusion:
jobs. This is done to ensure the availability of                    Providing the processor with a multi-
system resources to the job Manager component              node intelligent terminal-based system 600 times
running on node 1. Geopad is a user interface              faster than the previous system, with interactive
providing a user friendly working environment              software facilities including colour displays and
for Geovecteur Plus. It is used for submitting jobs        3D visualization capabilities pose many chal-
and for running interactive applications. Geopad           lenges to the geophysicist. With the increasing
could initially be invoked only from node 1. If            aspirations of management as well as interpret-
seismic jobs are run on node I the response time           ers for faster, better quality and more outputs the
gets impaired. When a user submits his job, he             need for change in work culture becomes impera-
can either specify the node or by default it can           tive. There has to be more intense interaction be-
go to any node.                                            tween the interpreter and processor to avoid ‘re-
                                                           doing’ the processes which get completed almost
         Fig. 7 shows a snapshot of user time and          too fast. Management of oil companies will need
system time during last 24 hours for print / plot          capabilities through online documentation and
server and the six nodes. There are not enough             Hypertext on the various parameters and inter-
batch jobs to keep all the nodes busy. Further             faces between the modules and packages. The
there is hardly any load on the print/plot server.         attitudes of processors have to undergo a sea
The load on node 1 is primarily because of inter-          change so as to achieve results through teamwork
active jobs and also because it is a logger ma-            and Knowledge Management for which the com-
chine. In this environment node I is heavily               pulsions were never so great in the past.
loaded compared to other nodes. With the 3D
data coming for processing, different data sets            Acknowledgements
had to be distributed to different processing                      The authors would like to sincerely thank
groups for processing. After conversion to                 Shri Y. B.Sinha, Director (Exploration) for pro-
Geovecteur format, interactive manual editing of           viding the new computer facilities at RCC,
seismic traces is the first job to ensure high qual-       Vadodara and also for according the approval to
ity output. With 6-8 users running interactive             publish this paper. Thanks are due to Shri
editing sessions on node I, It was observed that           N.C.Sharma, GGM (E) for Ills encouragement
the response was intolerably poor. At this stage,          and support. Sincere thanks are also due to Shri
there was a paradigm shift in the thought pro-             S. R.M. Karnawat, DGM (Prog.) and Shri A.V.K.
cess. It was apparent that the response time even          Suryanarayana, DGM (Geoph) for cooperation
with such a fast machine was not at all sufficient.        and support at every stage.The authors are thank-
Though the machine was fast enough for batch               ful to S/Shri H. Ghosh. C.V. Rao, Mul Raj, S.N.
processes like migration it was too slow for in-           Jaiswal, M.K. Sharma and D.N.Bain for provid-
teractive editing. Paging space was found to be a          ing necessary inputs.
                                                       6
                                                                            GEOHORIZONS Vol .6 No. 1
                                                                                 January 2001