=======
Minutes
=======

Meeting:  pcpd discussion
Date:     2 March 2006 14:00 GMT
Venue:    Teleconference

Attendance:
  UEDIN:  Kostas Kavoussanakis (KK), Alistair Phipps (AKP)
  DL:     Mark Leese (ML)

Actions
-------

[KK] Talk to Yee-Ting Li about status / history of pcpd

[KK] Ask Robert Harakaly about pcpd design docs / assessment of code quality

[ML] Determine exactly what needed from a token passing mechanism

[AKP] Download pcpd code, read paper, learn more about pcpd

Discussion
----------

Why do EGEE want pcpd? [KK]
- Tier 0 -> Tier 1 star
  - short term would be a benefit as would make e2emonit installable / gLite
    packageable; but we don't have pcpd in the short term
  - longer term, e2emonit not suitable (due to inability to fill pipes) and
    LHC-OPN solution preferred
  - ML suggests we go with James Casey's solution of downloaded crontabs to
    satisfy short-term need

- Tier 1 -> Tier 2 etc
  - e2emonit could be a reference implementation that countries can adopt if
    they don't have any other preference
  - pcpd needed for a plug-and-play solution

Why do Gridmon want pcpd? [ML]
- Crontabs on Tier 1-> Tier 2 / Tier 3 very messy, makes pcpd an option
- Gridmon configuration is star topology from RAL (Tier 1) to Tier 2 sites, then
  full mesh within Tier 2/3 groups (e.g. Scotgrid)
- With pcpd, could perhaps have a token for the star and a token for each mesh
- Note that RAL's Tier 0 connection is a dedicated link so do not have to worry
  about overlapping measurements

What do we need from pcpd? [all]
- inter-clique locking protocol would be needed in Gridmon (and other similar)
  case; apparently not included in current implementation
- implementation poor - Ratna's analysis shows security flaws.  ML thinks it
  might have been discarded in favour of crontabs even by EDG.  More
  investigation needed.  KK to discuss with Robert Harakaly and Yee-Ting Li.

DONM
- Friday 10/3 14:00GMT