Canonical Environment of an Enstore System:

5 Server PCs

0: Runs the file and volume databases – requires 4 large SCSI disks, software raid

1: Runs the pnfs servers – requires 4 large SCSI disks, software raid

2: Runs config, alarm, log, inquisitor server (+other related tasks)

runs the web server for the system

requires 2 large ide disks

3: Runs the console server and admin integrity checks

4: Runs the library manager and media changes

1 Mover PC per tape drive – scalable unit.

1 PC/drive is simpler to manage operationally than multiple drives and is not a big expense

Misc. Equipment necessary for operating a robust system

Power switches, Console monitor, Cyclades pods, 2 serial cables per node, blowers, etc…

All PCs are currently Intel Lancewood systems:

PIII-800 L440GX System: Intel L440GX MB,

2 x 800Mhz PIII Proc., (450 or better is fine)

1 GB ECC SDRAM MEM, (memory is so cheap now, why get less?)

Sys Disk 8 to 15 GB EIDE,

Mid Tower Case, (this is not crucial, 1-2U unit are fine)

Floppy Disk Drive

No Keyboard (ok if there is one)

No CDROM, (ok if there is one)

Minimum 300 W Power Supply.

They were last ordered on Sep 26. 2001 on PO #540646. ($1645/PC was last price.) We are aware that this hardware is end-of-life and are researching a stable replacement.

All nodes support Fast Ethernet. We use the on-board FE port. We are exploring replacing fast ethernet with gigabit ethernet on a new motherboard.

All tapes drives are the low voltage differential SCSI variety. We use the on-board SCSI device.

All PCs are running a 2.2.16 kernel with the following changes:

1. Multiple interface arp change – only arp on selected interface

2. Reboot after memory error

3. Large IDE support

4. Stop after a non-initiated scsi bus reset

5. Raid patches, on nodes that need them

6. St patches to allow operation of LTO rives, on nodes that need them

ACSLS, the STK robot controller and server, runs on a Sun node and is completely managed by STK, including backups of the system software and tape database. The Sun node is on a back-end network to prevent hacking or malicious tampering.

All PCs are monitored for temperature, node configuration, network compliance, etc.

The rest of the software is a hybrid of Fermi RH 5.2 and 6.2 and directly built from source code. We are currently investigating an upgrade to Fermi RH 7.1. (and beyond) to reach a new stable base-code platform.

All system software is backed up to remote disks and occasionally to tape. There is a clone procedure to bring new PCs into the system.

Enstore mover and server software is updated via CVS from the hppc cvs repository. When bugs are found, they are fixed, tested on a test stand and then put into production as rapidly as possible.

Enstore depends on these external products:

Python – we use our version, downloaded from python.org. The major changes are (1)No built-in libdb support, (2) static options to allow freeze process for encp to work

FTT – we use v2_10

SWIG – we use v1_1p5 (primarily for our interface to robot software and tape drives)

Other standard products used for producing documentation

Internally, Enstore uses BerkeleyDB for databases, HTMLgen for web pages, gadfly as a linter. All of these packages come with the enstore cvs distribution.