Hi,

 > > I'm a little bit hazy on the details, but there are some slides 
from the
 > > meeting here[1]:
 > > 
http://indico.cern.ch/getFile.py/access?contribId=8&sessionId=1&resId=1&materialId=slides&confId=106641
 > >
 > > The key part seems to be:
 > >
 > > - LHCb (and LCD) use tcmalloc.so
 > >  - glibc dependent
 > >    - May require recompilation when glibc changes
 > >  - Not part of SLC
 > > - Code run on RHEL/SLC 5.6 without recompiling tcmalloc.so broke

I was one of the ones on LHCb who discovered this, so yes I can confirm 
the SLC5.6 brought in a kernel update that breaks tcmalloc (actually, 
the bug is in tcmalloc, see below).

 > >
 > > It's not clear whether this is an OS bug at all, or someone doing
 > > something completely unreasonable in experiment software, but AIUI it
 > > affect RHEL5u6 as well, so it seems not to be an SL bug at least.

I would say a bug in tcmalloc, not SL or RHEL. See for instance

<http://code.google.com/p/google-perftools/issues/detail?id=305>

The fix is to move to google perftools 1.7

cheers Chris