1. Look at CRAK (http://www.ncl.cs.columbia.edu/research/migrate/crak.html) or
similar "Process Checkpoint/Restart" solutions.
I startted my UNIX career as a system admin of a VAX which worked a lot on doing
simulations of physical and chemical experiments, running for weeks and months doing mostly
number crunching.
Being able to checkpoint and restart processes (added to BSD 4.2 by Gil Shwed)
was a major life-saver (we had to take the system down to single user mode every Friday
for backups, not to mention the hardware crashes).


2. For reliability - look at having a box with dual power supplies. Also you might want to
keep a mirror raid (RAID 1) to accomodate for disk failures.
Speaking of disks - you should look at setting up the data disk under a managed volume,
(http://tldp.org/HOWTO/LVM-HOWTO/) so you can easely add disks later to that volume
without changing configuration (that's the way I was told it is done on Hamakor's mirrors site).
That might be a point in favour of SCSI or SATA, so you'll be able to utilise more disks better.
(some people might add you should have a dual NIC, with failover capabilities or somesuch,
but you are not running a web site are you? Though it would be difficult to access it without
a functioning NIC, try to price how much would it cost to setup a dual NIC compared to the
chance that you'll have to run over there and replace it if and when it fails).


3. I'd also recommand using ReiserFS for a filesystem - It's journaling, so no need to fsck when
booting after a crash and the amount of data loss in case of a hard crash is minimal. You could
achieve the same with other journaling filesystems if you have anything against ReiserFS, but I
personally am very satisfied with it).


4. I heard bad things about Western Digital. I understand there is a lot of religion about such
things (e.g. a company has a bad batch which spoils the name of an entire line or the entire
brand). Maybe you should dig sites like tomshardware.com or google about the brand and
model you choose, try to get yourself a "rocks/sucks" ratio about it. I have a Maxtor IDE drive,
80Gb/8Mb cache, 10K RPM.


5. A UPS with a good interface to Linux is almost a must, from what you are telling about
your requirements. Also setup SMART monitoring tools (http://sourceforge.net/projects/smartsuite/) to monitor your disks and lm-sensors
(http://secure.netroedge.com/~lm78/) to monitor the health of the rest of your hardware,
try to find out how your hardware (MoBo/BIOS/Disks) cooperate with Linux about this.


6. Write down a failure contingency plan - imagine failure scenarios and plan what you'll
do in each one of them. Keep it updated.


Cheers,

--Amos

Igor Zhidkov wrote:

Dear Guru's.

Please help me in the choice of the PC computer configuration for work under Linux (I intend to instal Fedora on it).

On this computer there will be a complicated scientific calculations performed: user-written programs, mostly C/C++, which run from several days to several weeks.


The critical usere demands are: -productivness -stable and robust work during high (close to or equal 100%) load -big hard disk space


For now I've chosen the following configuration:


CPU:  PENTIUM IV 2.8 GHZ 512CACHE 800 MHz HT
MotherBoard: ASUS P4P800 i865PE/DUAL DDR400/LAN
Memory : 2 X 512 MB DDR 400 MHz
Video Card: 128 MB ATI RADEON 9200 TVOUT AGP X8
HDD: WD 120GB/7200/8MB CACHE
DVD-ROM: ASUS
FDD 1.44": PANASONIC/SONY
Case: CASE for PENTIUM IV 2FANS


Please suggest me any changes which can increase stability of the the configuration during long-time 100% load. Or give an examples of PC configurations you've purchased recently.


Is Asus P4P800 motherboard good for long-time lasting hard use? Or it can be overheated, because it was not initially designed to give maximal stableness, but to provide a best production for limited time periods ?

My another thought is to buy the same configuration, but assembled by some of "brands" like HP - i.e. to order HP brand computer with the same configuration as shown above. As I know, brand name assembled PC computers usually are more stable - isn't it ?

The ideal case is to buy non-PC (SGI, Alpha, Sun) computer, but we're rather limited in money for such computers.


Thank you very much in advance for your suggestions, All !!!






=================================================================
To unsubscribe, send mail to [EMAIL PROTECTED] with
the word "unsubscribe" in the message body, e.g., run the command
echo unsubscribe | mail [EMAIL PROTECTED]



---------- Your email is protected by Mailshell ---------- To block spam or change delivery options: http://www.mailshell.com/control.html?a=blshp8b9gc0rxhgk_srox_llfpptvypmvy7j


ReturnPath.net http://rd.mailshell.com/ad481
Earn up to $3 for each of your friends who signs up with Mailshell! http://rd.mailshell.com/sp5




=================================================================
To unsubscribe, send mail to [EMAIL PROTECTED] with
the word "unsubscribe" in the message body, e.g., run the command
echo unsubscribe | mail [EMAIL PROTECTED]



Reply via email to