2001 Summer Student Lectures Computing at CERN Lecture

  • Slides: 27
Download presentation
2001 Summer Student Lectures Computing at CERN Lecture 3 — Looking Forwards Tony Cass

2001 Summer Student Lectures Computing at CERN Lecture 3 — Looking Forwards Tony Cass — Tony. Cass@cern. ch Tony Cass

Data and Computation for Physics Analysis event filter (selection & reconstruction) detector processed data

Data and Computation for Physics Analysis event filter (selection & reconstruction) detector processed data event summary data raw data batch physics analysis event reconstruction analysis objects (extracted by physics topic) event simulation interactive physics analysis Tony Cass 2

LEP and LHC Parameters Compared Tony Cass 3

LEP and LHC Parameters Compared Tony Cass 3

Evolution of CERN Computing Needs CPU Capacity 1997 -2002 120'000 Infrastructure Engineering Others 80'000

Evolution of CERN Computing Needs CPU Capacity 1997 -2002 120'000 Infrastructure Engineering Others 80'000 LEP 60'000 40'000 NA 48 NA 45 20'000 COMPASS LHC 2002 2001 2000 1999 1998 0 1997 CERN Units 100'000 year Tony Cass 4

Evolution of CERN Computing Needs Tape Storage 1995 -2000 1000 LHC Tera. Bytes 800

Evolution of CERN Computing Needs Tape Storage 1995 -2000 1000 LHC Tera. Bytes 800 COMPASS 600 NA 48 400 200 All LEP 0 1995 Others 1996 NA 49 NA 45 1997 1998 1999 2000 Year Tony Cass 5

10'000 9'000 8'000 7'000 6'000 5'000 4'000 3'000 2'000 1'000 0 Infrastructure Engineering Others

10'000 9'000 8'000 7'000 6'000 5'000 4'000 3'000 2'000 1'000 0 Infrastructure Engineering Others LEP NA 48 NA 45 COMPASS year Tony Cass 2006 2005 2004 2003 2002 2001 2000 1999 LHC 1998 1997 CERN Units Evolution of CERN Computing Needs CPU Capacity 1997 -2006 6

Evolution of CERN’s CPU requirements A different view… Estimated CPU Capacity required at CERN

Evolution of CERN’s CPU requirements A different view… Estimated CPU Capacity required at CERN K SI 95 5, 000 Moore’s law – 4, 000 some measure of the capacity technology advances provide for a constant number of processors or investment LHC 3, 000 2, 000 Other experiments Jan 2000: 3. 5 K SI 95 Tony Cass 2010 2009 2008 2007 2006 2005 2004 2003 2002 2001 2000 1999 0 1998 1, 000 7

Evolution of CERN Computing Needs Tape Storage 1995 -2006 14'000 10'000 8'000 LHC 6'000

Evolution of CERN Computing Needs Tape Storage 1995 -2006 14'000 10'000 8'000 LHC 6'000 4'000 Current Experiments 2'000 COMPASS 2006 2005 2004 2003 2002 2001 2000 1999 1998 1997 1996 0 1995 Tera. Bytes 12'000 Year Tony Cass 8

HELP! u The previous slides show that we have to cope with a dramatic

HELP! u The previous slides show that we have to cope with a dramatic increase in computing capacity before the start of LHC. – Can we afford it? – How many boxes are needed (i. e. can we manage the equipment)? u Fortunately, the price of computing equipment falls each year. Will this help us? Tony Cass 9

CPU and Disk Cost Predictions CPU Costs Disk Costs 350. 0 100. 0 300.

CPU and Disk Cost Predictions CPU Costs Disk Costs 350. 0 100. 0 300. 0 80. 0 250. 0 70. 0 CHF per GB CHF per CERN-Unit 90. 0 60. 0 50. 0 40. 0 30. 0 20. 0 les 150. 0 pasta low 100. 0 cms pasta high actual price 50. 0 10. 0 1996 200. 0 1998 2000 2002 2004 2006 0. 0 1996 year 1998 2000 2002 2004 2006 year Tony Cass 13

Disk Storage — The Bad News 1996 2000 4 GB 10 MB/s 50 GB

Disk Storage — The Bad News 1996 2000 4 GB 10 MB/s 50 GB 20 MB/s 1 TB I/O 250 x 10 MB/s 20 x 20 MB/s 2, 500 MB/s 400 MB/s Tony Cass 14

Tape Storage Estimates u ! u Although CPU and disk costs are expected to

Tape Storage Estimates u ! u Although CPU and disk costs are expected to decrease dramatically, current estimates are that the cost of tape storage and tape devices will fall by less than a factor of 2 over the next 8 years. – These are not commodity items! » Most tape use is for archive storage (write once, read never), not HEP like usage. » Who backs up their home PC? Tape storage and tape devices are expected to represent a significant fraction of the cost of computing for the LHC experiments, particularly for ALICE. Tony Cass 15

Which system architecture for LHC? Experiment Tape Storage Tera. Bytes of disks Tony Cass

Which system architecture for LHC? Experiment Tape Storage Tera. Bytes of disks Tony Cass Which of the different system architectures SMP Scalable Distributed is appropriate for the LHC experiments? 16

Networks and CPU load u High bandwidth commodity networks carry the baggage of their

Networks and CPU load u High bandwidth commodity networks carry the baggage of their low speed commodity origins. – The MTU 1 for Gigabit Ethernet is still the 1. 5 KB of Ethernet. » c. f. 64 KB for Hi. PPI. processing packets takes time—and, with Gigabit Ethernet, the packets come thick and fast. 1 MTU = Maximum. Transmission Unit Tony Cass 17

Can we build LHC Computing Farms? u Probably or almost certainly, depending on your

Can we build LHC Computing Farms? u Probably or almost certainly, depending on your level of optimism. u On the positive side – CPU and disk price/performance trends suggest that the raw processing and disk storage capacities will be affordable, and : B/s 9 – raw data rates and volumes look manageable 99 M » perhaps not today for ALICE. u But this does not mean it will be easy. 45 8! @ A 4 R N CD for 1 – Many, many boxes will be needed compared to today’s systems. – Building and managing coherent systems from such large numbers of boxes will be a challenge. : /s Tony Cass 0 B 0 M 0 2 90 ! @ lice A R CD for 18

CERN’s Network Connections 1 G b/s SWITCH RENATER National Research Networks Mission Oriented s

CERN’s Network Connections 1 G b/s SWITCH RENATER National Research Networks Mission Oriented s / b M IN 2 P 3 Link 2 s / b TEN-155: Trans 155 M 45 Mb/s WHO European Network at 155 Mb/s CERN 39/155 Mb/s TEN-155 155 M Public b/s KPNQwest (US) 1 Gb /s Commercial C-IXP Tony Cass 19

CERN’s Network Traffic June/July 2001 Incoming data rate TEN-155 4. 7 Mb/s 5. 5

CERN’s Network Traffic June/July 2001 Incoming data rate TEN-155 4. 7 Mb/s 5. 5 Mb/s 5. 2 Mb/s 40 Mb/s IN 2 P 3 14 Mb/s KPNQwest (US) 155 Mb/s 25 Mb/s CERN 45 Mb/s 20 Mb/s 21. 6 Mb/s Out 17. 9 Mb/s In Outgoing data rate Link Bandwidth 0. 1 Mb/s 2 Mb/s 0. 1 Mb/s 100 Mb/s ~10 TB/month in/out RENATER SWITCH Tony Cass 1 TB/month = 3. 86 Mb/s 1 Mb/s = 10 GB/day 23

Outgoing Traffic by Protocol May 31 st-June 6 th 1999 350 : e t

Outgoing Traffic by Protocol May 31 st-June 6 th 1999 350 : e t o N ata d 9 9 19 300 Giga. Bytes Transferred 250 200 150 Elsewhere USA Europe 100 50 0 ftp www X afs int rfio mail news other Total Protocol Tony Cass 24

Incoming Traffic by Protocol May 31 st-June 6 th 1999 350 : e t

Incoming Traffic by Protocol May 31 st-June 6 th 1999 350 : e t o N ata d 9 9 19 300 Giga. Bytes Transferred 250 200 150 Elsewhere USA Europe 100 50 0 ftp www X afs int rfio mail news other Total Protocol Tony Cass 25

European & US Traffic Growth Feb ’ 97 -Jun ’ 98 USA Start of

European & US Traffic Growth Feb ’ 97 -Jun ’ 98 USA Start of TEN-34 connection EU Tony Cass 26

European & US Traffic Growth Feb ’ 98 -Jun ’ 99 USA EU Tony

European & US Traffic Growth Feb ’ 98 -Jun ’ 99 USA EU Tony Cass 27

European & US Traffic Growth Mar ’ 99 -Jul ’ 00 USA EU Tony

European & US Traffic Growth Mar ’ 99 -Jul ’ 00 USA EU Tony Cass 28

European & US Traffic Growth. 1. Mar ’ 99 -Jul ’ 00 k 0

European & US Traffic Growth. 1. Mar ’ 99 -Jul ’ 00 k 0 c 0. e 2 n a t r e l a o t f D t o g 0 b n i 0 t a 0 a 2 pd r USA e g u n o h l t r o o n w e t r o a N ks r o EU tw e N Tony Cass 29

Traffic Growth Jun 98 - May/Jun 99 4 Total 8 Total 7. 00 3.

Traffic Growth Jun 98 - May/Jun 99 4 Total 8 Total 7. 00 3. 00 : e ot Incoming N ata d 9 9 19 EU 6. 00 5. 00 2. 00 Outgoing 4. 00 3. 00 2. 00 1 1 0. 00 ftp 8 7. 00 www X afs int rfio mail news other Total ftp 8 Other 7. 00 6. 00 5. 00 4. 00 www X afs int rfio mail news other Total US 4. 00 3. 00 2. 00 1 1 0. 00 ftp www X afs int rfio mail news other Total 0. 00 ftp Tony Cass www 30

Traffic Comparison 1998 1999 2000 u Two significant changes this year – Ba. Bar

Traffic Comparison 1998 1999 2000 u Two significant changes this year – Ba. Bar data flow from SLAC to IN 2 P 3 via CERN, and – increased use of ssh prevent a quantitative comparison of 2000 traffic patterns with those of previous years. u Qualitatively, however, physics data transfer is taking an increasing fraction of the overall traffic. – Ba. Bar traffic is one good example. – CMS have also relied on the network to transfer Monte Carlo data during their tests of their High Level Trigger algorithm. Tony Cass 31

Round Trip times and Packet Loss rates Round trip times for packets to SLAC

Round Trip times and Packet Loss rates Round trip times for packets to SLAC 98 es. 9 1 ur Fig [This is measured with ping; A packet must arrive and be echoed back; if it is lost, it does not give a Round Trip Time value. ] 5 seconds! Packet Loss rates to/from the US on the CERN link [But traffic to, e. g. , SLAC passes over other links in the US and these may also lose packets. ] Tony Cass 32

Looking Forwards—Summary u LHC demands for CPU and I/O capacity significantly exceed those of

Looking Forwards—Summary u LHC demands for CPU and I/O capacity significantly exceed those of the LEP experiments. – Fortunately, experiments such as COMPASS have intermediate requirements and allow us to study the problems before LHC startup. – CPU cost trends suggest we can afford distributed computing farms which provide adequate resources » but we have to start installing these in 2003/2004. u Software quality is a major concern for the LHC experiments. – Object Oriented techniques are being adopted. – This allows us to consider the use of Object Oriented Databases for data management and other commercial packages for analysis work. Tony Cass 39