Re: CycleCloud BigScience Challenge giving away ~8-hours on 30000 core cluster for research

From: Leonid Gibiansky Date: November 14, 2011 technical Source: mail-archive.com
We have just published a paper (L Gibiansky, E Gibiansky, R Bauer, in print, should appear in JPP in a week or so) that in particular evaluates parallel computing performance on a single 12-CPU Window 7 workstation. We found out that for the TMDD-QSS model in differential equations, "Use of parallel computations with 4 to 12 processors running on the same computer improved the speed proportionally to the number of processors with the efficiency (for 12 processor run) in the range of 85 to 95% for all methods except BAYES, which had parallelization efficiency of about 70%." Efficiency was defined as 100%*(single CPU run time)/(multiple CPU run time)/(number of processors) Roughly, the model run on 11 processors was 10 times faster than the single-processor run of the same model Leonid -------------------------------------- Leonid Gibiansky, Ph.D. President, QuantPharm LLC web: www.quantpharm.com e-mail: LGibiansky at quantpharm.com tel: (301) 767 5566
Quoted reply history
On 11/14/2011 9:54 AM, Bill Knebel wrote: > David, > > Some limited benchmarking results are listed at the bottom of this > email. It is also important to remember that there are ways that > cloud-computing helps beyond parallel NONMEM. Cloud-computing allows on > demand, user specific, clusters that can grow and shrink depending on > user requirements. This allows for rapid completion of large bootstrap > processes (500 - 1000 jobs) in the time it takes to run one, > non-parallel job. Users can also evaluate model variants simultaneously > with cluster size being increased or decreased as needed. Users do not > have to worry about competing for resources (compute cores) with other > users because cloud-computing clusters are user and/or project specific. > The performance gains are evident in parallel NONMEM jobs and single > modeling projects, but it is important to look beyond the simple > benefits of individual job/project speed-up and more towards the impact > of cloud-computing on the entire portfolio of modeling and simulation > projects in a given group or company. > > Bill > > Model 1 - ADVAN6, 1000 subjects, dual linear and non-linear elimination > CoresRuntime (hr) > 180.7 > 816.8 > 168.9 > 246.2 > 483.8 > 962.5 > > Model 2 - ADVAN6 - 70 subjects, PKPD model > CoresRuntime (hr) > 14 hr > 81.1 > 160.66 > 240.4 > 480.31 > > cores = number of compute cores (value of "NODES=" argument in NONMEM > pmn file) > > On Nov 9, 2011, at 9:28 PM, David Foster wrote: > > > I agree with Julia, > > > > Thanks for this Bill, but quantitative benchmarking would be very much > > appreciated. > > > > Regards, > > > > David > > > > On 10/11/11 9:55 AM, "Ivashina, Julia" <[email protected] > > <x-msg://692/[email protected]>> wrote: > > > > Bill, > > > > This is nice to hear about the speed improvements for NONMEM you > > received. > > > > Could you please describe gains in performance in a qualitative > > manner with model examples you used. > > I think everyone will benefit from such NONMEM 7.2 benchmarking. > > > > I posted a similar question in March but not many responded. > > > > Thanks, > > Julia > > > > -----Original Message----- > > From: [email protected] > > <x-msg://692/[email protected]> > > [mailto:[email protected]] On Behalf Of Bill Knebel > > Sent: Wednesday, November 09, 2011 5:03 PM > > To: [email protected] <x-msg://692/[email protected]> > > Cc: Jason Stowe; Nick Holford; William Bachman > > Subject: Re: [NMusers] CycleCloud BigScience Challenge giving away > > ~8-hours on 30000 core cluster for research > > > > Since there seems to be some questions around cloud-based NONMEM, > > we wanted to share our experience of utilizing parallel NONMEM in > > a cloud-based environment. We use the enterprise-licensed version > > of NONMEM(R) 7.2. to perform modeling and simulation, both serial > > and parallelized, on an AWS cloud infrastructure. Briefly, we used > > a combination of recent MPICH2 libraries, the Sun Grid Engine > > integrated ORTE parallel environment, and StarCluster to deploy > > parallel NONMEM(R) on linux AWS EC2 instances of up to 100 cores > > for estimation (which appears to be the current NONMEM(R) 7.2 > > limit for parallel computing), and a virtually unlimited number of > > cores for simulation or use of other open-source modeling tools > > (R, OpenBUGS). The gains in performance are very real, and > > probably the best we've experienced on any multi-user distributed > > computing infrastructure. > > > > Bill > > > > ~~~~~~~~~~~~~~~~~~~~~~~~ > > Bill Knebel, PharmD, PhD > > Principal Scientist II > > > > Metrum Research Group LLC > > 2 Tunxis Road, Suite 112 > > Tariffville, CT 06081 > > > > O: 860.735.7043 > > C: 860.930.1370 > > F: 860.760.6014 > > > > On Nov 4, 2011, at 3:06 PM, Bachman, William wrote: > > > > > Cloud computing on a system like this requires an unlimited > > enterprise license for NONMEM. Pricing is available from > > [email protected] <x-msg://692/[email protected]>. > > > > > > > > > William J. Bachman, Ph.D. > > > Director, Pharmacometrics R&D > > > Icon Development Solutions > > > 6031 University Blvd., Suite 300 > > > Ellicott City, MD 21043 > > > Office 215-616-8699 > > > [email protected] <x-msg://692/[email protected]> > > > > > > > > > > > > From: [email protected] > > <x-msg://692/[email protected]> > > [mailto:[email protected]] On Behalf Of Jason Stowe > > > Sent: Friday, November 04, 2011 2:12 PM > > > To: Nick Holford > > > Cc: [email protected] <x-msg://692/[email protected]> > > > Subject: Re: [NMusers] CycleCloud BigScience Challenge giving > > away ~8-hours on 30000 core cluster for research > > > > > > Hi Nick, > > > On Thu, Nov 3, 2011 at 8:59 AM, Nick Holford > > <[email protected] <x-msg://692/[email protected]>> > > wrote: > > > Jason, > > > > > > If I was a winner of your competition how would I go about using > > NONMEM? > > > How do you manage NONMEM licensing on you cloud cluster? > > > What NONMEM license fee would one have to pay for 30000 cores? > > > > > > Nick > > > Hi Nick, > > > Unfortunately, I can't speak to an individuals user's NONMEM > > license. We do have commercial clients running NONMEM on > > CycleCloud clusters at varying scales. A winner with appropriate > > licensing doesn't need to run 8 hours on 30k cores, you could also > > run for 24 hours on 10000 cores, etc. If your license relates to > > compilation vs. execution, or if you have a site-wide license, we > > can't say, so you'll need to look at your license to deal with that. > > > > > > Perhaps someone from Icon could talk about this kind of > > non-profit licensing? > > > > > > Thanks, > > > Jason > > > > > > > > > > > > On 2/11/2011 2:19 p.m., Jason Stowe wrote: > > > Hi all, > > > > > > Long time reader, first time poster. > > > > > > Cycle recently announced that we're giving away the equivalent of > > 8-hrs on a 30000 core CycleCloud HPC cluster for a grand prize > > winner in the CycleCloud BigScience Challenge. Up to five > > finalists will win 4 hours on 3000 core clusters as well. The > > Compute time is to benefit academic and non-profit researchers > > that are doing science which could benefit humanity. > > > > > > Some current applicants are in life sciences and CycleCloud > > currently supports reliable/large-scale clusters running NONMEM > > for Top 5 Pharma, so I wanted to welcome more researchers in the > > NONMEM community to enter. > > > > > > We describe the challenge and why we're excited about it here: > > > http://bit.ly/BigScience > > > > > > The application process is simply answering 4 questions (takes > > less than half an hour): State who you are, what is your research, > > why it is important, and how you currently run computation. The > > applicatiopn is available here: > > > http://cyclecomputing.com/big-science-challenge/overview > > > > > > So far, response has been great, and Inside HPC covered > > descriptions of some of the recent applications we've received: > > > http://insidehpc.com/2011/10/27/24209/ > > > http://blog.cyclecomputing.com/ > > > > > > Submissions are due by November 7th, so submit early and we hope > > to help some of you get some BigScience done quickly. > > > > > > Best, > > > Jason > > > > > > -- > > > > > > > > > ================================== > > > Jason A. Stowe > > > cell: 607.227.9686 > > > main: 888.292.5320 > > > > > > http://twitter.com/jasonastowe/ > > > http://twitter.com/cyclecomputing/ > > > > > > Cycle Computing, LLC > > > Leader in Open Compute Solutions for Clouds, Servers, and Desktops > > > Enterprise Condor Support and Management Tools > > > > > > http://www.cyclecomputing.com http://www.cyclecomputing.com/ > > > http://www.cyclecloud.com http://www.cyclecloud.com/ > > > > > > > > > -- > > > Nick Holford, Professor Clinical Pharmacology > > > Dept Pharmacology& Clinical Pharmacology > > > University of Auckland,85 Park Rd,Private Bag 92019,Auckland,New > > Zealand > > > tel:+64(9)923-6730 fax:+64(9)373-7090 mobile:+64(21)46 23 53 > > > email: [email protected] > > <x-msg://692/[email protected]> > > > http://www.fmhs.auckland.ac.nz/sms/pharmacology/holford > > > > > > > > > > > > > > > -- > > > > > > > > > ================================== > > > Jason A. Stowe > > > cell: 607.227.9686 > > > main: 888.292.5320 > > > > > > http://twitter.com/jasonastowe/ > > > http://twitter.com/cyclecomputing/ > > > > > > Cycle Computing, LLC > > > Leader in Open Compute Solutions for Clouds, Servers, and Desktops > > > Enterprise Condor Support and Management Tools > > > > > > http://www.cyclecomputing.com http://www.cyclecomputing.com/ > > > http://www.cyclecloud.com http://www.cyclecloud.com/ > > > > > >