PC's as computing farms: first experiences

1
PC's as computing farms first experiences

• Frédéric Hemmer

http//wwwinfo.cern.ch/pdp/pc/PCRD
2
Introduction

• PC RD started last autumn
• Goals
• Establish whether PCs can be used
• for data intensive computing
• in the context of NA45/NA48/Compass CDR
• as a alternative to large SMPs, CS/2s, etc.
• what are the limits
• Linux and Windows/NT
• Good demonstrator for LHC

3
Overall Status

• Hardware (reusable)
• 7 Dual PIIs _at_ 266 MHz
• 1 Alpha PWS _at_ 433 MHz
• 1 Sun Ultra 5 _at_ 270 MHz with HiPPI
• 100 GB disk
• Initial conclusions needed before NA48 data taking

4
Background

• Catherine Charbonnier, Fabien Collin, Soren
  Olofsson, Bernd Panzer, Cristoph Von Praun, Jes
  Sorenesen
• Performance (aimed at CDR for Compass)
• Network performance
• Disk performance
• Combination

5
Network performance Results

• PCs interconnected through 100 BaseT 3Com 3000
  switch
• Repeated with other H/W
• Half duplex behavior
• Block size does not matter
• Linux uses less CPU than NT
• ? Good unidirectional performance
• ? Disappointing CPU consumption on NT
• ? Disappointing bi-directional performance

6
PC to PC Network performance
7
Network performance issues

• Unexplained 0.5 MB/s observed with some eepro100
  versions on PCRD hardware, but OK on PCSF
• Recent DEC E'net boards with chipset gt 21140 give
  poor performance on Linux
• Surprising results PC/Alpha

8
PC/Alpha Network performance
9
Disk performance

• PCs connected to SEAGATE ST19171W using two
  Adaptec 2940 UW
• NT needs a lot of tuning (default behavior is to
  swap data out!)
• Block size, BIOS settings, EDO/FPM does not
  matter
• ? Poor performance
• ? Windows NT even worse
• ? Memory bandwidth is suspected

10
Disk performance

• Striping has no effect
• 1 stream 2 stripes 21 MB/s (22 max)
• 1 stream 3 stripes 21 MB/s (33 max)

11
Disk performance issues

• Memory bandwidth suspected
• Need to test with LX/SDRAM
• RISC PCI does not support variety of boards
• Combined disk/network performance even worse
  5-6 MB/s on Linux

12
Memory bandwidth (lmbench)
13
Initial conclusions (1)

• PCs can be used for low I/O tasks
• Confirmed with
• Simulation on PCSF
• NOMAD reconstruction on Linux
• NA45 reconstruction on NT
• PCs are not adequate now as disk servers

14
Initial conclusions (2)

• PC world diversity
• Large choice, manufacturer independent
• Difficulties
• EEPRO100/Base hardware problems
• DEC Tulip driver problems
• IDE Master/Slave problems on some HW
• VGA drivers for S3 Virge
• Lilo for disk gt 1024 cylinders
• PC world changes very fast, what is true today
  may not be tomorrow

15
PC Technology evolution in 97

• Pentium Pro ? Pentium II
• 50 raw performance increase
• but 50 cache performance reduction
• SEC ? new motherboards
• 440 FX ? 440 LX (SDRAM, AGP)
• Recent MBs ? embedded SCSI, Enet, VGA
• 100 Mbit Enet switches standard, 1000 Mbit
  arriving

16
Next steps

• Understand I/O bandwidth problem
• Measure LX/SDRAM based PCs
• Measure low-end RISC performance on
• Disk
• Network (HiPPI)
• Experiment with PC/HiPPI
• Management aspect has not been investigated
• Wake on LAN
• WfM

17
Questions

• Why PCs, what benefits drawbacks
• What are the parameters
• Should all components be PCs