partitioning HD for use of swap & for booting

[Alan Grossfield]

:As Greg and I both said, the system will use swap if it exists for
:various optimizations whether or not an application actually swaps,
:although it sounds like 2.4.x might be a bit weak still on the
:optimizations.  So far I haven't hit a lot of the instabilities
:personally, but neither have I hammered on any of my 2.4.x based
:systems.  Soon.

    My lab has hammered fairly hard on our nodes (950 MHz Athlons,
    ~500-600 MB memory) using Gaussian98 (single processor form) -- we
    found that upgrading to the 2.4.3 kernel (from 2.2.18, hand-compiled
    on top of RH 6.2) dramatically improved stability, and may have
    improved performance slightly.  The effect was quite dramatic --
    with Gaussian jobs running on perhaps 10-15 nodes, round the clock,
    I came to expect one or two nodes to hang (pings work, everything
    else apparently dead) per day, plus a few jobs would die here and
    there for no apparent reason.  Moreover, when Gaussian was streaming
    large amounts of data in and out of memory, the nodes would frequently
    become _very_ slow for a few minutes, then recover.  In the month or
    so since the upgrade, we've not had a single crash, and the nodes
    haven't gone unresponsive.  To be fair, we're not running quite as
    many jobs at the moment, but the failure rate has clearly plummeted.

    (these machines are set up with 2GB of swap)

Alan Grossfield
---------------------------------------------------------------------------
| ...reality must take precedent over public relations, for Nature cannot |
| be fooled.     Richard Feynman, Appendix to the Challenger Report       |
---------------------------------------------------------------------------