Now, this sure is a great list.. within hours after my posting, my
mailbox is suddenly flooded with solutions!
Thanks goes to:
anthony baxter <anthony.baxter_at_aaii.oz.au>
Gyula Szokoly <szgyula_at_skysrv.Pha.Jhu.EDU>
Christophe Colle <colle_at_krtkg1.rug.ac.be>
Michael Matthews <matthewm_at_sgate.com>
alan_at_nabeth.cxo.dec.com
JT McDuffie <mcduffie_at_kira.engr.sgi.com>
John Stoffel <john_at_WPI.EDU>
Allan Small <small_at_gidday.ENET.dec.com>
Jon Buchanan <Jonathan.Buchanan_at_ska.com>
Andrew Brennan <BRENNAN_at_HAL.HAHNEMANN.EDU>
"Jeffrey C. Ollie" <jeffo_at_worf.netins.net>
Joern Wilms <wilms_at_astro.uni-tuebingen.de>
Jim Belonis <belonis_at_dirac.phys.washington.edu>
William Flett <will_at_dcs.rhbnc.ac.uk>
Kurt Carlson <SXKAC_at_orca.alaska.edu>
"James D. Zelenka" <jz1j+_at_andrew.cmu.edu>
arun sanghvi <sanghvi_at_proto.wilm.ge.com>
"Paul E. Rockwell" <rockwell_at_rch.dec.com>
Anthony Jackson <ajackson_at_iii.com>
Phil Rand <prand_at_paul.spu.edu>
> 1.) Comparatively speaking, which has faster/higher performance, doubling
> the number of processors for an Alphaserver 2100 or doubling (increasing)
> the it memory size from 64 Mb to 128 Mb?
Most of those replies suggested that doubling/increasing memory would be
a better idea than doubling the number of processors.
The need for memory was also verified to be of greater importance over
cpu with the use of some normal system tools such as vmstat, iostat, ps,
etc. as well as public domain tools like top, monitor (from
gatekeeper.dec.com).
It was noted that the slow performance of the system was mostly spent on
paging memory in and out than it was actually computing anything -- ergo,
most of the applications are memory-bound than CPU-bound.
Paul Rockwell has this to say:
=============================
It depends on what the characteristics are of the program(s) that you're
trying to run.
If a single non-threaded program runs comfortably (with minimal paging)
in 64MB on a single CPU, then neither doubling the number of processors nor
increasing memory will help it.
If your program is not multi-threaded, but is causing a significant amount
of paging, then increasing memory is a good thing. Another CPU won't help.
If you have a mix of programs (with little paging activity), then adding the
second processor would probably be a big win.
If you're running a mix of programs and paging the brains out of the system,
you have a severely underconfigured system. Add memory first, then consider
a second CPU if you need the throughput.
In general, add memory when you have a memory shortage. Adding memory
decreases paging activity and makes more room in the unified buffer
cache. Add the CPU after resolving memory issues to simultaneously
execute multiple processesor threads, which should improve throughput. If you
need CPU cycles, add CPU. But don't expect to add a CPU to solve problems
caused by insufficient memory.
> 2.) Is it possible to set up the finger daemon of a site in such a way that
> when someone fingers such host, it would always show no user that is
> logged in INSPITE the fact that there really is someone logged on?
Some suggested that disabling the finger daemon by commenting the finger
entry at the /etc/inetd.conf and send a -HUP to the inetd process.
Others suggested writing a short finger daemon would do the trick, while
others suggested picking finger source codes from many GNU, Linux or
FreeBSD sites and hack it to ones' preferences.
Bonn
:)
Received on Mon Nov 20 1995 - 13:12:27 NZDT