RAL Site report

1
RAL Site report

• John Gordon
• ITD
• October 1999

J.C.Gordon_at_rl.ac.uk
2
Summary

• Linux Farm
• NT Farm (Monday)
• Suns for BaBar (Friday)
• Disk and Tape
• Security(Wednesday)
• Y2K

3
Linux

• Linux in use in most parts of CLRC
• Formed a user group to share experiences
• For central HEP systems more Linux cpu power than
  any other system

4
Hardware Configuration

• Twenty built to measure PCs
• SuperMicro Dual Motherboard (with SCSI)
• Two Pentium II 450
• 10GB 5400rpm IDE HDA
• 256MB ECC memory
• 100Mbit Ethernet (tulip or Intel)
• Cheap graphics card
• Usually run without monitor - (BIOS must allow
  this)

5
Pounds per MHz
6
(No Transcript)
7
Hardware Costs Per Dual System

• Dual CPU System 1450
• Shelving 10
• Power 30
• Network 50
• Software 0
• Cost Per Pentium 450 CPU 900 (inc VAT)

8
Cloning

• Presently trivial but labour intensive
• System image created by dd onto SCSI tape
• memory resident Linux system run from floppy
  (Toms Root and Boot)
• dd from tape to system disk
• Need to become smarter!
• Kickstart?
• Drive Image (or similar software)?
• Any other suggestions?

9
Software

• Redhat 5.2 (kernel 2.0.36)
• ARLA (Free AFS Software)
• Generic NQS 5.4 - Free but not recommended -
  evaluating commercial products
• Mainly Fortran 77 - therefore use g77 compiler
  (egcs 1.1.1). Some C
• autorpm for system updates

10
Summary

• Procurement Lessons
• System Monitoring
• Redhat 5.2 needed several changes
• Problems

11
Procurement (lessons)

• Procurement was run as two tenders 4 months
  apart. Hardware is (and will continue to be) a
  moving target.
• Detailed (but not detailed enough) specification
  of all components.
• Watch out for warranty terms! Need to pin down
  details.
• Acceptance tests vital (ours are still evolving).
  Not all H/W delivered identical (as required)

12
System Monitoring

• Service needs to be highly reliable.
• lm_sensors (Hardware monitoring).
• System monitoring scripts, check filesystem
  occupancy, load average, batch system status
• Operations staff notified via our automated
  operations system (SURE)
• System logs spool to system logger for security
  monitoring.
• What about SMART, ECC, SERR/PERR?

13
Tuning/Tweaks (for Redhat 5.2)

• Large disks need geometry setting explicitly
• Memory autosizing is unreliable. Set explicitly
  at boot (mem256M)
• NFS (user level) client is poor (better in kernel
  2.2). Set rsize/wsize explicitly (also in 6.0?).
• NFS implementation buggy - mount with option
  timeo0. (Directory cache timeout)
• disk dma transfer mode on
• Insufficient VFS inodes - Increase

14
Current Problems

• Hardware/OS reliability is fair - 1 break every
  40 PC weeks. Not as good as HP-UX
• NQS is buggy - needed to eyeball/hack code
• ARLA is buggy - occasional cache hangs
• Process accounting is buggy - accounting files
  become corrupted
• Most of this does not impact the users - managed
  by monitoring/workarounds.

15
Plans

• Need to move to Redhat 6.0 (or 6.1)
• Disk mirroring for interactive service?
• Next expansion will be late Autumn. Probably
  based on dual Pentium 600.
• Possibly further expansion early next year
  (probably Pentium)
• Further expansion 2H2000 when Multi-processor AMD
  Athlon systems will be extremely interesting
  possibility.

16
Disk

• Always growing
• 1.25TB general user disk servers
• 4.5TB for BaBar
• Plan to test an IDE server

17
Tape

• 30TB IBM3590 in 3494 robot
• STK robot idle - considering upgrade to Eagle
  drives.