Chandra Observers,
This message is of interest only to people who use the CIAO software
on a Mac platform.
As you know, the CXC has not yet approved CIAO for use under the
recently-released Mac operating system (10.5.1, "Leopard"). Those of
you who are using CIAO under Leopard anyway (or are thinking of doing
so) will be interested to know about a serious performance problem we
have encountered at our site. We have found that when dmcopy is run
on data that is accessed through NFS (network file system) then its
execution time is ~50 times longer than normal.
NFS most commonly comes into play when you are accessing a file that
is physically on a different computer system. However, please note
that in some computing environments local disks can be accessed
through paths which involve NFS, and thus suffer this performance
problem. For example, on my machine the one of the disks is
physically mounted at /Volumes/Bulk2/ but can also be accessed via the
NFS path /bulk/tecumseh2/. If I cd to a Chandra data directory via
the first path then dmcopy will operate normally, however if I cd via
the second path then files will be accessed through NFS and dmcopy
will be very slow.
This effect is not due to general network delays---other processes
such as fcopy and cp ran quickly over NFS in our testing. We know
nothing about the root cause of this very odd phenomenon. We have no
idea if this problem is pervasive among Leopard machines, or is
isolated to our site. We do not know if other CIAO commands are
affected.
If anyone out there experiences this problem and finds a solution I
hope they will take the time to contact chandra-users with advice.
Regards,
Patrick Broos
ACIS Team @Penn State
This archive was generated by hypermail 2b29 : Tue Feb 12 2013 - 01:00:12 EST