September 2014 - Synergy-team - mailman.cs.vt.edu

Your submission to the OpenACC Hackathon
by Foertter, Fernanda S. 08 Sep '14

08 Sep '14

Dear Team, Thank you for your submission to OLCF's inaugural Hackathon. All the entries were fantastic and we struggled quite a bit with choosing applications. Unfortunately, we only had enough mentors for 5 applications, and we were unable to chose your team. But given the interest and excitement, I foresee more Hackathons in the near future. Please consider applying for a Directors Discretion with our center to continue your GPU porting project. You can find details about access to Titan via the link below. The application is a short webform. Under "How did you hear about the OLCF and this discretionary allocation award process?" please answer "OLCF Hackathon." I will personally advocate for your Team's access to our resources. https://www.olcf.ornl.gov/support/getting-started/olcf-director-discretion-… If you would like to receive announcements about upcoming training opportunities at OLCF, you can subscribe to our ccs-training email list. To subscribe, simply send an email to ccs-training-join(a)email.ornl.gov<mailto:ccs-training-join@email.ornl.gov> I look forward to seeing you at a future OLCF training event. Please don't hesitate to stay in touch. Happy Computing! Fernanda -------------------------------------------------------- Fernanda Foertter National Center for Computational Sciences Oak Ridge National Laboratory O: 865-576-9391 F: 865-241-2850 foertterfs(a)ornl.gov<mailto:foertterfs@ornl.gov>

1 0

Re: [Synergy-team] Hackathon team submission from team Synergy@VT
by Foertter, Fernanda S. 05 Sep '14

05 Sep '14

Dear Team Synergy@VT, The bad news is we do not yet have a decision. The good news is that the applications were so many and so good that we need a few more days to consider the entries. Our decision announcement has moved to Monday. Apologies for the delay! Happy Computing! Fernanda -------------------------------------------------------- Fernanda Foertter National Center for Computational Sciences Oak Ridge National Laboratory O: 865-576-9391 F: 865-241-2850 foertterfs(a)ornl.gov<mailto:foertterfs@ornl.gov> On Aug 29, 2014, at 4:24 PM, OpenACC Hackathon: Synergy@VT <synergy-team(a)cs.vt.edu<mailto:synergy-team@cs.vt.edu>> wrote: Team Name Synergy@VT Email synergy-team(a)cs.vt.edu<mailto:synergy-team@cs.vt.edu> Tell us about your application Application Name: mpiBLAST (http://www.mpiblast.org/), 2004 R&D 100 Award. (Note: We are not confined to the above application. We have two other applications of interest as part of larger project in large-scale computing and big data, namely ABySS and Ray, which we briefly discuss at the end of this application.) Application Purpose: The BLAST family of sequence search algorithms serves as the foundation for much biological research. It searches for similarities between a query sequence and a large database of DNA or amino-acid sequences. Similarities between the new sequence and a gene of known function can help identify the function of the new sequence. Other uses include phylogenetic profiling, genome alignment, and understanding gene function and evolution. LOC: 1.2M Languages and Libraries: MPI and NCBI BLAST Current performance characteristics: Using a novel combination of database segmentation and query segmentation (the latter which is the "embarrassingly parallel" way of parallelizing BLAST), mpiBLAST delivers both better throughput and faster response time per query. It has been shown to scale up to 65,536 processors of an IBM BlueGene/L supercomputer at KAUST. Where does it run? CPUs only. Expected performance gains? The expected performance gain will depend on the granularity of parallelization that we will be able to tackle during the "Hackathon." If we parallelize at the coarse-grained level (as in our GPU-BLASTP algorithm from IPDPS'11), then the performance gain will be modest, e.g., 2x; if at the more sophisticated and fine-grained level (as in our cuBLASTP algorithm from IPDPS'14), then the performance gain will be a bit higher, e.g., 3x to 4x). The performance gain will also be dependent on the NCBI BLAST toolkit. Tell us about your team The team consists of members from the Synergy Laboratory, which is led by Professor Wu Feng, the PI for mpiBLAST. Of the team members, Jing Zhang, a PhD student of Professor Feng, is the most familiar with mpiBLAST due in part to his experience with related projects, e.g., cuBLASTP (IPDPS'14). He also is an experienced CUDA programmer but does not have any experience with OpenACC. The remaining members have a range of experience in parallel computing but are not familiar with mpiBLAST. 1. Jing Zhang, 3rd-year PhD student, CUDA and life sciences expertise. Recent publication on cuBLAST in IPDPS'14. 2. Kaixi Hou, 3rd-year PhD student, CUDA expertise and OpenCL and MPI experience. Recent publication on graph traversal on Xeon Phi. 3. Rubasri Kalidas, 2nd-year MS student, OpenCL experience. No publications yet. Female member of the team. 4. Yilong Jin, 2nd-year PhD student, OpenCL experience. Working on an OpenCL version of MapReduce for CPU+GPU. 5. Islam Harb, 1st-year PhD student but with 5+ years of industry experience in parallel hardware and architecture, CUDA experience. U.S. citizen. 6. Xiaodong Yu, 2nd-year PhD student, CUDA expertise but no experience with OpenACC or MPI. CUDA GPU publications as part of his MS. If selected, I understand at least 2 team members will travel to the hackathon on Oct 27th-31st * I understand What computing facilities does this application run on? Local clusters, desktops, and HPC centers. At one point, mpiBLAST was being distributed as part of Linux distributions. Please check skills your team members have experience with, if any * Developing in C/Fortran * Developing in C++ * Compiling applications * Profiling applications * Debugging applications using debuggers * GPU Architecture * GPU Programming (OpenACC, CUDA, OpenCL) * Linux environment (logging in remotely, using terminal, basic commands) * Experience with multiple compilers (GCC, PGI, Cray etc) Is there anything else you'd like to mention or ask us? If HPC = supercomputing, e.g., Titan, then the team's experience in HPC is limited at best, particularly with respect to MPI, where Kaixi Hou is the only one with adequate MPI experience. If HPC = massive parallelism within a node via CUDA, OpenCL, or OpenMP, then the team's experience is fairly extensive. In addition to the primary email contact, synergy-team(a)cs.vt.edu<mailto:synergy-team@cs.vt.edu>. Please feel free to contact Professor Wu Feng at feng(a)cs.vt.edu<mailto:feng@cs.vt.edu> or 540-231-1192 if you have any questions.

1 0

Re: [Synergy-team] Hackathon team submission from team Synergy@VT
by Foertter, Fernanda S. 01 Sep '14

01 Sep '14

Dear Synergy Team, Thank you for your submission. We will announce teams on Sep 5th. Please email me directly if you have any questions! Happy Computing! Fernanda -------------------------------------------------------- Fernanda Foertter National Center for Computational Sciences Oak Ridge National Laboratory O: 865-576-9391 F: 865-241-2850 foertterfs(a)ornl.gov<mailto:foertterfs@ornl.gov> On Aug 29, 2014, at 4:24 PM, "OpenACC Hackathon: Synergy@VT" <synergy-team(a)cs.vt.edu<mailto:synergy-team@cs.vt.edu>> wrote: Team Name Synergy@VT Email synergy-team(a)cs.vt.edu<mailto:synergy-team@cs.vt.edu> Tell us about your application Application Name: mpiBLAST (http://www.mpiblast.org/), 2004 R&D 100 Award. (Note: We are not confined to the above application. We have two other applications of interest as part of larger project in large-scale computing and big data, namely ABySS and Ray, which we briefly discuss at the end of this application.) Application Purpose: The BLAST family of sequence search algorithms serves as the foundation for much biological research. It searches for similarities between a query sequence and a large database of DNA or amino-acid sequences. Similarities between the new sequence and a gene of known function can help identify the function of the new sequence. Other uses include phylogenetic profiling, genome alignment, and understanding gene function and evolution. LOC: 1.2M Languages and Libraries: MPI and NCBI BLAST Current performance characteristics: Using a novel combination of database segmentation and query segmentation (the latter which is the "embarrassingly parallel" way of parallelizing BLAST), mpiBLAST delivers both better throughput and faster response time per query. It has been shown to scale up to 65,536 processors of an IBM BlueGene/L supercomputer at KAUST. Where does it run? CPUs only. Expected performance gains? The expected performance gain will depend on the granularity of parallelization that we will be able to tackle during the "Hackathon." If we parallelize at the coarse-grained level (as in our GPU-BLASTP algorithm from IPDPS'11), then the performance gain will be modest, e.g., 2x; if at the more sophisticated and fine-grained level (as in our cuBLASTP algorithm from IPDPS'14), then the performance gain will be a bit higher, e.g., 3x to 4x). The performance gain will also be dependent on the NCBI BLAST toolkit. Tell us about your team The team consists of members from the Synergy Laboratory, which is led by Professor Wu Feng, the PI for mpiBLAST. Of the team members, Jing Zhang, a PhD student of Professor Feng, is the most familiar with mpiBLAST due in part to his experience with related projects, e.g., cuBLASTP (IPDPS'14). He also is an experienced CUDA programmer but does not have any experience with OpenACC. The remaining members have a range of experience in parallel computing but are not familiar with mpiBLAST. 1. Jing Zhang, 3rd-year PhD student, CUDA and life sciences expertise. Recent publication on cuBLAST in IPDPS'14. 2. Kaixi Hou, 3rd-year PhD student, CUDA expertise and OpenCL and MPI experience. Recent publication on graph traversal on Xeon Phi. 3. Rubasri Kalidas, 2nd-year MS student, OpenCL experience. No publications yet. Female member of the team. 4. Yilong Jin, 2nd-year PhD student, OpenCL experience. Working on an OpenCL version of MapReduce for CPU+GPU. 5. Islam Harb, 1st-year PhD student but with 5+ years of industry experience in parallel hardware and architecture, CUDA experience. U.S. citizen. 6. Xiaodong Yu, 2nd-year PhD student, CUDA expertise but no experience with OpenACC or MPI. CUDA GPU publications as part of his MS. If selected, I understand at least 2 team members will travel to the hackathon on Oct 27th-31st * I understand What computing facilities does this application run on? Local clusters, desktops, and HPC centers. At one point, mpiBLAST was being distributed as part of Linux distributions. Please check skills your team members have experience with, if any * Developing in C/Fortran * Developing in C++ * Compiling applications * Profiling applications * Debugging applications using debuggers * GPU Architecture * GPU Programming (OpenACC, CUDA, OpenCL) * Linux environment (logging in remotely, using terminal, basic commands) * Experience with multiple compilers (GCC, PGI, Cray etc) Is there anything else you'd like to mention or ask us? If HPC = supercomputing, e.g., Titan, then the team's experience in HPC is limited at best, particularly with respect to MPI, where Kaixi Hou is the only one with adequate MPI experience. If HPC = massive parallelism within a node via CUDA, OpenCL, or OpenMP, then the team's experience is fairly extensive. In addition to the primary email contact, synergy-team(a)cs.vt.edu<mailto:synergy-team@cs.vt.edu>. Please feel free to contact Professor Wu Feng at feng(a)cs.vt.edu<mailto:feng@cs.vt.edu> or 540-231-1192 if you have any questions.

2 1

Test Message
by Wu Feng 01 Sep '14

01 Sep '14

1 0

Fwd: Synergy-team post from foertterfs@ornl.gov requires approval
by Wuchun Feng 01 Sep '14

01 Sep '14

Email from "Foertter, Fernanda S." <foertterfs(a)ornl.gov> for ORNL Hackathon. (Note: Hao is eligible.) This is a manual re-send as the set-up for the mailing list appears to not be working. ---------- Forwarded message ---------- From: <synergy-team-owner(a)cs.vt.edu> Date: Fri, Aug 29, 2014 at 4:27 PM Subject: Synergy-team post from foertterfs(a)ornl.gov requires approval To: synergy-team-owner(a)cs.vt.edu As list administrator, your authorization is requested for the following mailing list posting: List: Synergy-team(a)cs.vt.edu From: foertterfs(a)ornl.gov Subject: Re: Hackathon team submission from team Synergy@VT Reason: Post by non-member to a members-only list At your convenience, visit: https://cs.vt.edu/mailman/admindb/synergy-team to approve or deny the request. ---------- Forwarded message ---------- From: "Foertter, Fernanda S." <foertterfs(a)ornl.gov> To: "OpenACC Hackathon: Synergy@VT" <synergy-team(a)cs.vt.edu> Cc: Date: Fri, 29 Aug 2014 20:26:43 +0000 Subject: Re: Hackathon team submission from team Synergy@VT Dear Synergy Team, Thank you for your submission. We will announce teams on Sep 5th. Please email me directly if you have any questions! Happy Computing! Fernanda -------------------------------------------------------- Fernanda Foertter National Center for Computational Sciences Oak Ridge National Laboratory O: 865-576-9391 F: 865-241-2850 foertterfs(a)ornl.gov<mailto:foertterfs@ornl.gov> On Aug 29, 2014, at 4:24 PM, "OpenACC Hackathon: Synergy@VT" < synergy-team(a)cs.vt.edu<mailto:synergy-team@cs.vt.edu>> wrote: Team Name Synergy@VT Email synergy-team(a)cs.vt.edu<mailto:synergy-team@cs.vt.edu> Tell us about your application Application Name: mpiBLAST (http://www.mpiblast.org/), 2004 R&D 100 Award. (Note: We are not confined to the above application. We have two other applications of interest as part of larger project in large-scale computing and big data, namely ABySS and Ray, which we briefly discuss at the end of this application.) Application Purpose: The BLAST family of sequence search algorithms serves as the foundation for much biological research. It searches for similarities between a query sequence and a large database of DNA or amino-acid sequences. Similarities between the new sequence and a gene of known function can help identify the function of the new sequence. Other uses include phylogenetic profiling, genome alignment, and understanding gene function and evolution. LOC: 1.2M Languages and Libraries: MPI and NCBI BLAST Current performance characteristics: Using a novel combination of database segmentation and query segmentation (the latter which is the "embarrassingly parallel" way of parallelizing BLAST), mpiBLAST delivers both better throughput and faster response time per query. It has been shown to scale up to 65,536 processors of an IBM BlueGene/L supercomputer at KAUST. Where does it run? CPUs only. Expected performance gains? The expected performance gain will depend on the granularity of parallelization that we will be able to tackle during the "Hackathon." If we parallelize at the coarse-grained level (as in our GPU-BLASTP algorithm from IPDPS'11), then the performance gain will be modest, e.g., 2x; if at the more sophisticated and fine-grained level (as in our cuBLASTP algorithm from IPDPS'14), then the performance gain will be a bit higher, e.g., 3x to 4x). The performance gain will also be dependent on the NCBI BLAST toolkit. Tell us about your team The team consists of members from the Synergy Laboratory, which is led by Professor Wu Feng, the PI for mpiBLAST. Of the team members, Jing Zhang, a PhD student of Professor Feng, is the most familiar with mpiBLAST due in part to his experience with related projects, e.g., cuBLASTP (IPDPS'14). He also is an experienced CUDA programmer but does not have any experience with OpenACC. The remaining members have a range of experience in parallel computing but are not familiar with mpiBLAST. 1. Jing Zhang, 3rd-year PhD student, CUDA and life sciences expertise. Recent publication on cuBLAST in IPDPS'14. 2. Kaixi Hou, 3rd-year PhD student, CUDA expertise and OpenCL and MPI experience. Recent publication on graph traversal on Xeon Phi. 3. Rubasri Kalidas, 2nd-year MS student, OpenCL experience. No publications yet. Female member of the team. 4. Yilong Jin, 2nd-year PhD student, OpenCL experience. Working on an OpenCL version of MapReduce for CPU+GPU. 5. Islam Harb, 1st-year PhD student but with 5+ years of industry experience in parallel hardware and architecture, CUDA experience. U.S. citizen. 6. Xiaodong Yu, 2nd-year PhD student, CUDA expertise but no experience with OpenACC or MPI. CUDA GPU publications as part of his MS. If selected, I understand at least 2 team members will travel to the hackathon on Oct 27th-31st * I understand What computing facilities does this application run on? Local clusters, desktops, and HPC centers. At one point, mpiBLAST was being distributed as part of Linux distributions. Please check skills your team members have experience with, if any * Developing in C/Fortran * Developing in C++ * Compiling applications * Profiling applications * Debugging applications using debuggers * GPU Architecture * GPU Programming (OpenACC, CUDA, OpenCL) * Linux environment (logging in remotely, using terminal, basic commands) * Experience with multiple compilers (GCC, PGI, Cray etc) Is there anything else you'd like to mention or ask us? If HPC = supercomputing, e.g., Titan, then the team's experience in HPC is limited at best, particularly with respect to MPI, where Kaixi Hou is the only one with adequate MPI experience. If HPC = massive parallelism within a node via CUDA, OpenCL, or OpenMP, then the team's experience is fairly extensive. In addition to the primary email contact, synergy-team(a)cs.vt.edu<mailto: synergy-team(a)cs.vt.edu>. Please feel free to contact Professor Wu Feng at feng(a)cs.vt.edu<mailto:feng@cs.vt.edu> or 540-231-1192 if you have any questions. ---------- Forwarded message ---------- From: synergy-team-request(a)cs.vt.edu To: Cc: Date: Fri, 29 Aug 2014 16:27:04 -0400 Subject: confirm 5f053a7bf1b1414b744699a7158be4149e1d6385 If you reply to this message, keeping the Subject: header intact, Mailman will discard the held message. Do this if the message is spam. If you reply to this message and include an Approved: header with the list password in it, the message will be approved for posting to the list. The Approved: header can also appear in the first line of the body of the reply.

1 0