New Page

Phobos Computers at RCF

Phobos has the following machines at RCF for computing:

Number	Processor	Clock Speed	Manu.	RAM/CPU MB	Local Disk/Node (GB)	O/S	IP Address
69	Dual Pentium III	1000 MHz	IBM	256	144 (2x72 SCSI)	RH6.1	rcrs4075->rcrs4138 rcas4011->4015
46	Dual Pentium III	800 MHz	VA Linux	256	110 (4x30IDE)	RH6.1	rcrs4030->rcrs4074 rcas4010
29	Dual Pentium III	450 MHz	VA Linux	256	110 (4x30IDE)	RH6.1	rcrs4001->rcrs4029
5	Quad Pentium-Pros	200 MHz	ASL	256	26 (2x18SCSI)	RH6.1	rcas4001->rcas4005
4	Dual Pentium-II's	400 MHz	IBM	256	50 (2x30IDE)	RH6.1	rcas4006->rcas4009

The machines are grouped into 2 broad categories.

Central Reconstruction Server (CRS) type machine
Machines dedicated to running phobos general reconstruction/pass0 jobs. Can access HPSS directly(like file system), runs via special RCF+Phobos scripts only. Only one account, phobreco, can run jobs on these machines. Has basic queuing system.
Central Analysis Server (CAS) type machines
Machines dedicated to running phobos user analysis jobs. Can not access HPSS directly. Access via LSF batch queueing system (except for 10 interactive machines).

Phobos Constraints:

Because Phobos requires more computing power than $$ can buy, efficient use of computing resources is essential. To this end, the current set up of machines, has each machine capabile of being either a CRS or CAS machine. Of course, with flexibility comes complexity, the following describes the more complicated(but more efficient) setup than was previously implemented.

Basic Machine Groupings:

CRS Queue Number	5	4	3	2	1
LSF Queues	phcrs_lo	phcrs_hi phcrs_med phcrs_lo	phcas_hi phcas_med phcas_lo		phslow_hi phslow_med phslow_lo
Number Of Nodes	32	32	23	22	29
Clock Speeds(MHz)	1000	1000	800	800	450
IP addresses	rcrs4107-138	rcrs4075-4106	rcrs4052-74	rcrs4030-51	rcrs4001-4029
Purpose	Primary purpose for CRS machine type processing. (Reconstruction) If necessary,or CRS underutilized, CAS machines.	Primary purpose for CRS machine type processing/overflow. If CRS requires more power, will use these machines, otherwise available to CAS users. CRS has priority. Generally used for MC/template generation if not used for reconstruction	Primary purpose for user analysis + MC/template generation. If required for CRS processing can be used (generally unavialable to CRS)	Primary purpose for user analysis. If required for CRS processing can be used (generally unavialable to CRS)	Primary purpose for user analysis. If required for CRS processing can be used (generally unavialable to CRS)
Authorization Required to use as CRS	Yes, std permission.	Yes, special permission	Yes, exceptional circumstances	Yes, exceptional circumstances	Yes, exceptional circumstances
Authorization Required to use as CAS	Yes, exceptional circumstances	Yes, general circumstances	No	No	No

Details of the CRS queues:

There are 5 queues:

Queue=5 is the main std CRS queue, where the reconstruction is run
Queue=4 is the overflow queue from the CRS queue=5. MC and generation jobs can run here, via LSF, if reconstruction not using. Also used for repassing over data.
Queue=3 is an overflow queue from queue 4. From CRS point of view, mainly used for major repasses. This queue shares its processing power with the CAS queues. The CRS machines are allowed to have 2 running jobs at any time. They have priority with nice=0. The CAS machine queues are less nice.Generally no CRS jobs should be running on this queue.
Queue=2 is the queue that shares its processing power with the CAS queues. It is the overflow, from queue=3. The CRS machines are allowed to have 2 running jobs at any time. They have priority with nice=0. The CAS machine queues are less nice. Generally no CRS jobs should be running on this queue.
Queue=1 is the CRS queue that is only used during special circumstances. Generally these nodes are marked unavaible to the CRS nodes.

For how to use the CRS farm click here.

Details of the CAS queues:

There are many queues on "CAS". The queues, phXXX_YY, are defined by the group of machines(XXX) that they use(of single speed), and by the length of time(YY) they can run. The full description of the queues is given.

LSF Queue Name	CRS Equivalent Queue No.	Machine IP addresses	Time/user	No. Jobs /user	Nice	Queue Priority
phcrs_hi	4	rcrs4075-4106	6 hrs	10	5	40
phcrs_med	4	rcrs4075-4106	24hrs	50	10	30
phcrs_lo	4+5	rcrs4075-4138	unlimited	unlimited	20	10
phcas_hi	2+3	rcrs4030-4074	6hrs	10	1	40
phcas_med	2+3	rcrs4030-4074	24 hrs	50	6	30
phcas_lo	2+3	rcrs4030-4074	unlimited	unlimited	11	10
phslow_hi	1	rcrs4001-4029	6 hrs	10	0	40
phslow_med	1	rcrs4001-4029	24 hrs	50	5	30
phslow_lo	1	rcrs4001-4029	unlimited	unlimited	10	10
phobos_int_phast	Special	rcas4011-4015	1.5 hrs	3	5	50
phobos_int	Special	Interactive Machines (rcas4001-10)	1.5 hrs	2	5	50

Queue priority: Highest number, is the queue whose jobs go onto available nodes first.

For how to use the LSF queue click here.

Interactive Machines:

Slower machines, rcas4001-4005(200MHz)+rcas4006-4009(400MHz)+rcas4010(800MHz)
To login, at gateway, type
> ssh rcas40XX [ENTER] (where XX=00-10)

To run interactively on the interactive queues (phobos_int_phast, phobos_int) for fast compiles/quick processing, type
>bsub -q phobos_int -Is tcsh [ENTER] (or phobos_int_phast)
this gives an interactive shell that you can use on the faster machine. Run your phat compile, or quick test and then log out.

How to run a job via LSF on a specific node:
bsub -m "hostname" script
(for LSF/bsub options do man bsub etc)

This page is maintained by Nigel George.

Last Edited: Friday, September 21, 2001 17:03