CCR Autunno 2008 Gruppo Server Michele Michelotto at

CCR Autunno 2008 Gruppo Server Michele Michelotto at pd. infn. it CCR Oct-08 LNF michele michelotto - INFN Padova

Quanti SPECINT • SI 2 K – the nickname for SPEC CPU Int 2000 benchmark – Came after Spec 89, Spec Int 92 and Spec Int 95 • SI 2 K CERN – Measured with gcc compiler and “-O 2 –f. PIC –pthread” optimization – Multiparallel to express machine rate • SI 2 K LCG – SI 2 K CERN + 50% – Used to express computing power as seen by WLCG CCR Oct-08 LNF michele michelotto - INFN Padova 2

Quanti SPECINT • SI 2006 – SPEC pubblishes result only for post 2006 processor • SI 2006 rate – SPEC pubblished results running on all cores • SI 2006 CERN – SI 2006 measured with “gcc –O 2 –f. PIC –pthread”, multi-parallel CCR Oct-08 LNF michele michelotto - INFN Padova 3

SPEC rate vs parallel CCR Oct-08 LNF michele michelotto - INFN Padova 4

FZK Measurement • In 2001 SPEC with gcc was 80% of the average pubblished data • In 2006 the gap was much wider 80% CCR Oct-08 LNF michele michelotto - INFN Padova 5

Nominal SI vs real SI • FZK uses for tender SI 2 K with FZK tuning (gcc-high) and add 25% to “normalize” to year 2001 • CERN and FZK Proposal to WLCG: use SI 2 K with CERN tuning (gcc-low) and add 50% to normalize • Run n copies in parallel – Where n is the number of cores in the worker node – To take in account the drop in performance of a multicore machine when fully loaded. CCR Oct-08 LNF michele michelotto - INFN Padova 6

WLCG SI 2 K How-to • Run SI 2000 with gcc 3, 32 bit, with CERN flags – gcc –O 2 –f. PIC –pthread –m 32 • Run N copies of this SI 2000 in parallel as the N number of cores • Sum all the results • Add 50% – This is the SI 2 K of one machine • Sum over all the machines of your farm CCR Oct-08 LNF michele michelotto - INFN Padova 7

Exercise • Compute the WLCG official rating of a farm with 224 Dell Blade M 1000 e 2 x 5420 – Number of cores/server: 8 – SI 2 K gcc-low: 10218 – Total SI 2 K: 2289 k. SI 2 k + 50%: –Total WLCG SI 2 K: 3433 k. SI 2 k CCR Oct-08 LNF michele michelotto - INFN Padova 8

Performances CCR Oct-08 LNF michele michelotto - INFN Padova 9

Price / Perfomances Bold: 2008 NB: k. SI 2 K LCG Euro/(k. SI 2 k+50%) CCR Oct-08 LNF michele michelotto - INFN Padova 10

Best buy • Intel processor give best performance at intermediate clock • AMD processor give best performances at lower clock • Compare with recent Tier 2 tender CCR Oct-08 LNF michele michelotto - INFN Padova 11

New processor • Xeon 7400. Last version of Penryn for MP market – The 5400 core in the 7300 market – First six core from Intel • Nehalem DP “Gainestown” – Q 4 2008 – 40% faster in Int, 70% faster in FP, quad core, two thread per core (8 logical cpu), integrated DDR 3 – 5502: 1. 86 GHz, 4 MB(L 3), 60 W, 188$ 5570: 2. 96 GHz, 8 MB(L 3), 95 W, 1386$ • Opteron DP “Shangai” – Feb 09 – Double L 3 caches (6 MB) of Barcelona, still DDR 2 – 2372 HE: 2. 1 GHz, 55 W 2386 SE, 2. 8 GHz, 105 W CCR Oct-08 LNF michele michelotto - INFN Padova 12

HEPIX group • A group with people from the major lab (CERN, FZK, DESY, RAL, INFN, JLAB, TRIUMF) after IHEPCCC request • And people designated from experiments (CMS, ATLAS, ALICE, LHCB) • Several machine – (lxbench cluster) at CERN – Harpertown and Barcelona INFN PD – Harpertown in Desy CCR Oct-08 LNF michele michelotto - INFN Padova 13

Results • SI 2 K CERN, SI 2006 CERN but also SFP 2006 CERN in good agreements with experiments results • Before July: Internal Agreement to use SI 2006 CERN • After July: push to use CPU 2006 CPP in order to have more FP benchmark – About 10% FP like observed with Perfmon on lxbatch at CERN CCR Oct-08 LNF michele michelotto - INFN Padova 14

SPEC CPP • • 471. omnetpp 473. astar 483. xalancbmk 444. amd 447. deal. II 450. soplex 453. povray Integer tests Floating Point tests • We finally take aritmetic mean of the geometric mean of integer part and geomtric mean of FP part. • It’s wrong! We should take the geometric mean of the 7 tests, but difference is small if int and fp performances are very similar CCR Oct-08 LNF michele michelotto - INFN Padova 15

How to compute CPP CCR Oct-08 LNF michele michelotto - INFN Padova 16

What’s next? • Make a table of performance of currents processor with the new benchmark • Understand Nehalem wrt multithread • John Gordon group (or Gonzalo Merino? ) inside WLCG CCR Oct-08 LNF michele michelotto - INFN Padova 17