Summary of Panel Session A2


Three talks:
Astushi Manabe of KEK  (dolly++)

dolly++ is an installation tool which makes use of a ring-based
virtual network topology to install nodes.  Basically each node just
contacts its nearest neighbor for the install packets it needs
so that 100 nodes are not contacting the install server all at once.

Philip Papadopoulos of San Diego (ROCKS)

ROCKS is a tool to help users develop their own customized kickstart
file to get the configuration they want.  At UCSD it is used to
reinstall a node whenever the configuration is in doubt.  An HTTP
install server is used for reinstall.

Jarek Polok (CERN) talking about Euro Grid WP4, fabric managment.
More generalized talk--the installer from this package will use
LCFG and ASIS--installing the OS as well as the many CERN-specific
applications.

-----------------------------------------------------

Questions--do you buy-in installation services?
Some buy hardware installation, none buy software installation.

Questions--do you buy pre-configured systems or build your own
configuration...
Most have a special configuration that is supplied by the vendor

Question--Do you upgrade the entire cluster at one time or in rolling mode
Most upgrade the entire cluster at one time...and upgrade usually
consists of wiping the nodes clean and a fresh install of the next
version.

Question--do you perform formal acceptance or burn-in test.
FNAL does, none of the others present did.  NERSC requires vendor
burn-in before shipping.

Key tools mentions:
Dolly++, ROCKS (described above)

ekvm--functionality of KVM switch across network
PXE--way to initialize a node booting across net.

CTCS--system torture program from VA Linux

BIOS Writer--way to write BIOS on INTEL board--but have to use with care

LCFG--incorporates VA's system imager

-----------------------------------------
Issues brought up--policy on when to reinstall
Scalability of install servers
non-universality of RPM.





------------------------------------------------------------------
Steven C. Timm (630) 840-8525  timm@fnal.gov  http://home.fnal.gov/~timm/
Fermilab Computing Division/Operating Systems Support
Scientific Computing Support Group--Computing Farms Operations