Kors Bos, NIKHEF, Amsterdam

1
Introduction

• Kors Bos, NIKHEF, Amsterdam
• GDB meeting _at_CERN, Oct.4, 2006

2
From the last meeting

• Minutes of the Sept. meeting at the GDB wiki page
  
• https//twiki.cern.ch/twiki/bin/view/LCG/GridDeplo
  ymentBoard
• Comments ?
• Thank you Jeremy Coles !!
• Minutes, Summary and Actions now also on GDB wiki
• https//twiki.cern.ch/twiki/bin/view/LCG/GridDeplo
  ymentBoard
• Actions from Sept meeting
• https//twiki.cern.ch/twiki/bin/view/LCG/GdBAction
  List

3
Changes and apologies

• Estonia will join as a CMS Tier-2 site
• Mario Kadastik will be at the November GDB
  meeting
• Apologies received
• Richard Gokieli (Poland)
• Roger Jones (UK)
• Ruth Pordes (US)

4
Dates for 2006

• Oct 4 GDB _at_CERN IT Aud.
• Problem with eIRG meeting in Helsinki on Oct.4-5
• Nov 8 GDB _at_CERN Bld 60-6-002
• Note unusual place only phone, no vrvs
• Possibly pre-GDB meeting on Nov.7 on Storage
  Classes again
• Dec 6 GDB _at_CERN IT Aud.

5
Other Dates for September 2006

• Oct.18 FTS Admin Day at SARA, Amsterdam
• Please sign up
• Oct.24 C-RRB meeting at CERN
• Closed meeting
• I dont claim this to be the full list !

6
GDB meetings in 2007

• To avoid confusion in 2007,
• the GDB is
• always on the first Wednesday of the month

Decide before the end of 2006

• January 10
• February 7
• March 7
• April 4 in Prague
• May 2
• June 6

• July 4
• August 1 (probably not)
• August 31 in Vancouver
• September 5
• October 3
• November 7
• December 5

The CERN IT Auditorium has been reserved those
days
7
GDB meeting _at_Triumf in 2007

• Hi Kors,
• I forgot that I was supposed to contact you
  about having a GDB meeting at TRIUMF in
  association with CHEP'07. In talking to Randy
  Sobie, we thought the following would be good
• - Friday, Aug.31 GDB at TRIUMF
• - Saturday/Sunday, Sept. 1-2 LCG (Tier-2)
  meeting in Victoria
• - Sept.3-7 CHEP'07 in Victoria
• This would mean that people would have to fly in
  to Vancouver on Thursday. We can arrange for a
  bus from TRIUMF to the airport after the meeting
  on Friday so people can get to Victoria easily
  (30 minute flight). Alternately, we could have
  the bus drive everyone to Victoria via the ferry,
  which would be much more scenic. Let me know if
  this works we can work out the details later.
• Cheers, Mike Vetterli

8
Again and again and again CERN migration to
SLC4

• Change to SL4 64-bits needs a new version of VDT
• Distribution still expected in October? 31st of
  October?
• Some LXPLUS and BATCH nodes already on SL4
• SLC4 full rollout at CERN still in the autumn ?
  
• Well planned transition period
• When do the sites change over ?
• When do the LHC experiments change over ?
• When do we stop the WLCG support for the SL3
  version ?
• May have different timing for EGEE

9
Super Computing 2006

• SC06 is in Tampa, Florida Nov. 11-17
• http//sc06.supercomp.org/
• There is a first time EGEE booth
• CERN participates with CalTech
• Participating UK (RAL), FNAL/SLAC, ANL,
  CalTech/CERN, NL(SARA/NIKHEF), FZK, NDGF(?),
  EGEE, OSG. Alice/ROOT, who else?
• New WLCG flyer in the make!
• WLCG poster?, EGEE poster(s)?
• IC demo based on Google Earth!
• Network demo? Mona Lisa? LHCOPN flyer or poster?
• Shall we set up a mailing list?

10
Big Files

• Small files are
• Takes more time to transfer (tcp/ip protocol)
• Takes more timespace to store, specially on tape
• Takes more timeeffort to retrieve from tape
• You get too many of them
• Bigger files are
• More appropriate for raw data transfer storage
• T0-T1 raw data data distribution
• Useful for dataset definition grouping of files
• All ESD data belonging to one luminosity record
  (Atlas)
• and ESD data goes to all Tier-1s but also to
  many Tier-2s
• But bigger files are also
• More difficult to copy into your (AFS) home
  directory .. Quota!
• More difficult to copy onto your laptop
• What are the limiting factors? Applications?

11
Maximum Files sizes

• File sizes mostly limited to 1 or max 2 GB
• Reactions to my email suggest larger is possible
• May be as big as a tape 200 GB now
• May span several tapes 1TB
• Not 50 GB yet leaves too much dead wood
• Increases with increased disk sizes, i.e. time
• Preferred seems now 10 GB
• May grow to 50 GB beyond 2010
• Is there a plan?

12
Network test suite

• OPN will be described by David Foster this
  afternoon
• We know the T0-T1 network and rates quite well
• We know the T1-T1 network and rates also well
• We know less well the T1-T2 network and rates
• Design a plan to test T1-T2 connections
• E2e routing and bandwidth measurement
• The same tool to be used at each site
• Compare measured bandwidth with requirement
• Develop a test suite
• based of iperf ?
• Launched as part of SAM
• Advice on a solution if there is a problem
• Action on Don Petravnick, KB, Bruno Hoefft
• Any help or suggestion much appreciated

13
Fail-over solutions

• At least one Tier-1 site is down almost all the
  time
• After repair has to take in raw data at twice the
  rate
• Not the best condition to try a new fix
• If a 10 T1 is down the others could also take 1
  more
• Assuming there are 11 T1s and they are all equal
• Requires a democratic raw data distribution
• True for all experiments?
• Who distributes the raw data? Experiments or
  CERN?
• Relaxes hardware purchases
• 10 overcapacity OK but factor 2 is expensive
• Rebalancing done at the end of the year

14
End of Introduction

• Kors Bos, NIKHEF, Amsterdam
• GDB meeting _at_CERN, Oct.4, 2006