Hub/Switch Installation Guide

Chapter 3 System Preparation
200 September 2002 HPSS Installation Guide
Release 4.5, Revision 2
❖ ControlHostname & DataHostname- The controlhostname shouldreference
the network interface over which the requestsare sent fromthe SSandPVLto
the Mover. It is typically an ethernet address (although this is certainly not
required), since that is a reasonably low-latency network and this interface
does not require very much bandwidth.
The data hostname is used by the passive side Mover (the disk side for an
intra-HPSS transfer between disk and tape, and the read side for disk-disk or
tape-tape), when it creates a listen port for the actual data. This is typically a
higher bandwidth interface than the control hostname – usually something
like gigabit ethernet, HiPPI or the SP switch. Note that this setting has no
bearing on Mover to client transfers, as the network interface on the Mover
used for that transfer is solely determined by the operating system’s routing
table based on the client address.
❖ Executable Name - Verify that the named executable supports the specific
optionsthat arerequiredforthisMover. Onthedevice end,makesurethat the
correct device driver interface is enabled (e.g., _ssd for the IBM Atape driver
for 3490E/3590/3580, _omi for the Greshem/OMI driver for StorageTek
drives, _dd2 for Ampex DST drives, _bmux for the IBM BMUX attached
3490s). On the client side, make sure that the correct client access mechanism
is enabled (e.g., _gpfs for local file support).
❖ Encryption Key Set - This is not critical – but the encryption key should be set.
This should ensure that there is proper security placed on the TCP/IP
connection to the Mover (the one used by the SS and PVL).
➢ Migration Policy List
❖ Runtime interval - Verify that this is a sane value. In most cases, for disk
migration you don’t want this to be going off 5 minutes after the last run
completed – nor do you likely want it not going off for 5 days. For tape
migration, a larger value is likely more appropriate (but we don’t have much
experience here at a production site…).
❖ LastReadInterval(tape)-Thisshouldideallybebasedontheworkload–how
longafterafilehas beenreadisitquiteunlikelytobe accessedagain?Sincewe
don’treally getthislevel ofunderstandingveryoften andsince wedon’thave
anyrealexperience– Iwouldhazarda guessthata muchlargervaluethan the
“Last Update Interval” for disk would be appropriate.
❖ Last Update Interval (disk) - This should ideally be based on the workload –
how long after a file has been written is it quite unlikely to be written again?
Since we don’t really get this level of understanding very often, some other
ruleof thumbistypically used–I think30or 60minutesisusually reasonable.
A lower value might beusedifasiteisconcernedaboutgettinga tapecopyof
thedatasooner afterthedata iswritten(worstcase isthatwemight need tore-
migrate the data later – if it is subsequently updated).
❖ Target FreeSpace- If this is not100%, whynot? For a disk migration,one goal
is to get the data backed up to tape so in general we should migrate every file
that meets the other migration criteria.