1. ‘Computer power’ budget for the CERN Space Charge Group Alexander Molodozhentsev for the CERN-ICE ‘space-charge’ group meeting March 16, 2012 LIU project

2. LIST of the CERN machines  Under consideration by the ICE space- charge group … in frame of the LHC Injectors Upgrade Project … PS Booster  4 people PS  3 SPS  2 LEIR  2 ‘RCS’ study  1 Alexander Molodozhentsev

3. Main items to study ‘SHORT-term’ tracking … Convergence study  PSB, PS, SPS, LEIR Injection process  PSB, PS, SPS, LEIR Optimization of the machine operation (‘working point’ scan)  PSB, PS, SPS, LEIR Implementation the machine imperfections  all … MT extraction, bunch splitting mechanism  PS ‘LONG-term’ tracking …  simulations for a full-cycle of the machine for the ‘optimized’ set of the main parameters  all … Current status Alexander Molodozhentsev

4. Computational tools PTC-ORBIT code…  … compiled for the ‘lxplus’ machine …  … ‘batch’ runs … by using the ‘lxbst2001…2010’ Alexander Molodozhentsev CPU time depends on (2&1/2D model):  Number of the Space Charge Nodes around the machine  Number of the transverse mesh points for the Poisson solver  Number of the macro-particles to represent the ‘real’ beam  computational approach to simulate the space charge kick … should be OPTIMIZED for each machine to minimize the required CPU time per turn without artificial effects …

5. CERN PS Booster / CPU time Alexander Molodozhentsev NO ‘Space charge’ module WITH ‘Space charge’ module PIC & FFT with chamber N SP = 199  ??? N mesh = 128x128 N bin = 128 … should be optimized ! 2 sec3 sec Lxplus (interactive) NOT Processors  Processes !!! … always 8 processors ONLY !

6. CERN PS / CPU time Alexander Molodozhentsev 12 sec WITH ‘Space charge’ module PIC & FFT with chamber N SP = 70 N mesh = 128x128 N bin = 128 … should be optimized ! Lxplus (interactive)

7. Required budget (1) … o the ‘code setting’ optimization …  convergence study … Alexander Molodozhentsev ‘short-term’ tracking … at least one synchrotron period   1000 turn PS Booster   3sec  1000 turns   1 hour … at least 15 runs   15 hours PS   12 sec  1000 turns   3.4 hours … at least 10 runs   34 hours ASSUMPTION: ‘waiting’ time is ZERO !

8. o ‘Tune-scan’ analysis (FULL SCAN)  multi-particle tracking during a few synchrotron periods …  2’000 turns …  Q SCAN  0.5 …  Q STEP  0.025  (20  20) points  400 points Required budget (2) … Alexander Molodozhentsev PS Booster   3sec  2000 turns   1.7 hours (one point) … 400 points  1.7 h   28 days … PS   12 sec  2000 turns   6.7 hours (one point) … 400 points  6.7 h   111 days … ASSUMPTION: ‘waiting’ time is ZERO !

9. TOTAL CPU budget for the CERN Space Charge Group for 2012 year (8 calendar months  6000 h)  50…60 processors / RUN   5’000 ÷ 8’000 hours of the dedicated cores of the ‘lxplus’ cluster … Required budget (3) … Alexander Molodozhentsev

10. Additional comparison:  ‘batch’ regime …  mpich1 vz mpich2  lxplus (engpara) vz cs-ccr-beabt1 Engpara  … a set of 40 batch worker nodes have been equipped with low-latency 10Gb ethernet cards for batch runs of MPI applications. The nodes are Viglen CPU servers with 8 core Intel "Nehalem" L5520 chips and 48Gb of memory. Cs-ccr-beabt1  … no information in Internet v.0330

11. CERN PS / CPU time Lxplus (batch) 12.06 sec 14.35 sec Machine: lxbsu2311

12. CERN PS / CPU time N p = 22