Multicore Computational Center
Multicore Computational Center
As we come to the end of 2008 which corresponds to the 2nd quarter of the second year of the IBM award establishing the Multicore Computing Center, it is useful to take the time to reflect on where we are, what have been our major accomplishments and what we hope to accomplish in 2009.
I would first like to point out that within the UMBC CSEE department, we now have added a number of highly competent, professional, dedicated faculty staff and a host of research assistants who are making numerous contributions to the field of computational science. In addition, we have brought in a number of grants, faculty awards and collaborative industrial partnerships that clearly are a direct outcome of the IBM support to the MC2. I would like to recognize the permanent supported staff consisting of Prof.’s Ye. Yesha. J. Dorband, S. Zhou, Ya. Yesha, and RA’s D. Chapman, P.Phuong, C. Lohr, S. Kannan, N. Walia, T. Celestin, C. Wu and U. Obinna, in this year end report. I would also like to recognize some of our other associated faculty staff and students such as Prof.’s T. Finin, A. Joshi, M.Olano and their students K.Kishallay, N. Golpayegani, Z. Syed, and J. Martineau.
The accomplishments have been both in terms of providing computer systems resources (i.e. H/W and S/) as well as in computer applications. With the very recent arrival of this years IBM SUR award of equipment, we will be adding 4 QS 22s with 32GB ram, 5QS 22s with 8 GB of ram, 3 JS22 with 32 GB ram and 1JSS 22 with 8 GB to our computing facility which will double the available compute power. This additional combination of the QS22 and JS22 each with 32GB of ram will be the basis of future experiments in moving towards a hybrid architecture. Moreover, as a result of the work of Dorband and Lohr, our on line disk storage system has also doubled from 7 TB to ~15TB of raid 5 disk. . We expect the QS22 and Js22 roll outs in late January.
The hardware system has been stabilized under Dorband’s reconfiguration of the system. He has reconfigured the JS series bluegrit so that14 J S 21s running on bluegrit are consistent with all the JS20 nodes with upgraded Linux: JSxx are Yellow Dog and QSxx is Fedora 8. Installation of SDK 3.1 can now run on all nodes and installation of FORTRAN. (32 JS21, 32 JS20 14 QS20) is now complete.
The software system now seems headed in the right direction as the staff implemented such software packages as MPI, xlf Fortran, Hadoop, Apache Tomkat and Axis 2 and a host of other software systems needed for the kinds of applications our government, industry partners and university research faculty need.
In terms of the computational application contributions, we have published or presented more than a dozen papers at conferences and journals in the short time frame of our existence. Our students have won multi- year fellowships (D. Chapman) and been selected at major conferences for submitting the best student paper (P.Phuong). Our faculty has gained more than ½ dozen government grants as a direct result of our computational application studies.
A major impact of our applications research is the accomplishments achieved from developing the SOAR system. Our work has resulted in many papers and presentations at major national and international scientific conferences such as the AGU, IGARRS and WEBIST. We have completed the development of all the components needed to generate a 40 year outgoing long wave fundamental radiance data record on demand at arbitrary spatial and spectral resolutions. This included the completion of a software gridding package called GRIDDER that can ingest multi-satellite data and efficiently grid it employing David Chapman's raycasting algorithm on the Cell cluster. We have shown Gridder to accept five types of instruments – AIRS including the visible data, MODIS, HIRS 3, VTPR and are working on the SIRS instrument.
Our goal in this project is to create the longest continuous global satellite climate dataset available on line to all researchers. No such data product exists and we hope that this will be the first such data set that can be used to study climate changes over the past four decades. This has never been done. Such an intensive processing requirement would not be possible without the enhanced bluegrit computer configuration provided by IBM. We have installed HADOOP, a cloud computing paradigm using 30 JS20 nodes and conducted comparative performance tests with 1, 4 16, 30 nodes on various spatial data resolutions. . Parallelism scaled well and was a significant improvement over our traditional parallel system of allocating a days worth of satellite data to a given processor.
As a result of the capabilities we are developing, new challenging opportunities are occurring for MC2 to test these systems on other NASA and NOAA data sets. In one case, we have met with the NASA Ozone processing team mgr and staff who face a compute data intensive challenge to deliver by Dec. 1, 2009 an operational processing system to NOAA capable of generating ozone profiles from the instrument OMI using up to 500 of its spectral channels in real time. Their current system would take 300 days to process one day requiring a speed up of more than 100-fold. They plan to make available to MC2 their benchmark algorithms to evaluate our H/W and S/W systems for their problem. They have 5 years of OMI data so they also want to do a catch up in processing. This is a problem for a petaflop computer.
With Shujia Zhou on board now, we are working on Global Earth Observing Systems (GEOS 5) to port it to Cell cluster. Shujia has optimized the solar radiatiopn part of the GEOS 5 model on the Cell and is now working to execute the dynamics portion of the model. Shujia reported status at AGU.
A related application involves porting the Weather Regional Forecasting WRF model to the Cell cluster for performance evaluations. Working wit Prof. Arking of JHU and his student Po Ma, we have made substantial progress with this development. Solved parallel version of netCDF reader which was a roadblock affecting the development of this parallel version
In the area of Medical Imaging, we have received 80 microscopic cellular data sets from NIST to use to evaluate our RHSEG algorithm on bluegrit. We have completed the processing over night and delivered the results to NIST for comparative evaluations.
Under the leadership of Prof. Yelena Yesha, significant progress is being made on research in the area of multicore computing for business intelligence. Yelena gave an invited talk at NAEC 2008 in Octrober, 2008. SAP awarded 2 PhD fellowships in this area. A proposal submitted in collaboration with Prof. Kalpakis for consideration by IBM in the area of S-Streaming proposal was accepted and work is planned to start after the winter break.
Dr. Shujia Zhou gave an invited lecture at Cell workshop at CASCON2008, Toronto Canada.
I am also pleased to announce that MC2 was awarded our first industry Partnership through I/UCRC for an “intrusion Detection Project (pattern searches)”. Charles Lohr will be assigned to support this effort under mentorship by John Dorband. In addition, Dr. Dorband and Ya Yesha had their grant in developing a TB suffix array from LTS/NSA extended for another year.
In the spring semester, Prof. Dorband and Zhou have developed the curricula for a course they will be offering on “Parallel Computing with emphasis on the Cell”.
Phuong Nguyan has submitted her proposal to IBM CAS Toronto for a PhD Fellowship.
The major upcoming event in the next quarter will be conducting an NSF sponsored I/UCRC Planning Meeting scheduled for Jan. 21-23, 2009 in conjunction with Ga. Tech, UCSD and U/MN. We are looking forward to establishing through this mechanism, in collaboration with other Cell sites at Ga. Tech, UCSD and U/MN, partnerships with a broad constituency of industry and government agencies to explore and improve the effectiveness and productivity of multi-core computing applications.
Sincerely,
Milt Halem, Director