Hello Everyone
I'm thinking about doing the following:
Starting with one Netbackup master server, one media server and one Scalar 1000
tape library using AIT tapes, I'm thinking of establishing five volume pools per
tape library (additional media servers and tape libraries will be added but at
different physical sites). The volume pools will be designated with a site
identifier and the suffices alpha, beta, full-alpha, full-beta, and baseline.
I intend to do weekly full backups to the volume pool full, alternating daily
differential and cumulative backups to the pools alpha and beta. For example:
day 1 full backup to pool full-alpha or beta on alternating weeks
day 2 differential backup to pool alpha
day 3 cumulative backup to pool beta
day 4 differential backup to pool alpha
day 5 cumulative backup to pool beta
day 6 differential backup to pool beta
day 7 cumulative backup to pool alpha
The reversal of the pools on day 6 and day 7 is to balance the load between the
two pools. The start of this seven day cycle will be selected to be two days
prior to the expected peak days of the week for the application.
When the second media server/tape library is installed at another site, volume
pool beta and pool full-beta will be moved to the second tape library but both
tape libraries will have the same master server.
There will be a baseline class without a backup frequency, backups in this class
will be started manually when needed to preserve a set baseline and tapes
written in this pool will be removed from the tape library for storage. This
class will not have multiplexing set in order to have the data stored in a
fashion most suitable to being imported at a later date. This class will most
likely be used selectively when production upgrades are made to the server farm.
Tape duplication when made will be made from the respective alpha to the
respective beta pools.
I do not intend to use multi-streaming as there will be plenty of clients to
parallize against, minimal interference with the application is a primary
concern, and I would prefer the log messages indicative on a per client basis
rather than a per filesystem basis.
When a third and fourth media server/tape library is installed, a second master
server will be established and the the master servers will be co-ordinated via
global data manager. At that point client responsibilities will be divided
between the two clusters, hopefully in a fashion that respects the actual WAN
connections.
The clients will be distributed throughout the country, and will be well
connected via ATM to each site. I'm not certain as to the actual data sizes
that we're talking about as the developers seem to be a bit pre-occupied but the
client count will likely grow well into the thousands.
I'm also thinking that perhaps the media servers should have OC-12 or multiple
OC-3 connections to the respective site switches in order to avoid the need for
jumbo packets just to tune a gigabit ethernet. The clients are likely frame
based NIC's converted to cell via the respective switches and routers.
Any comments, suggestions, ideas about suitable hardware choices? We're mostly
a Sun shop as far as servers is concerned but we would probably use either
Compaq or Dell for PC servers if the Intel architecture makes more sense. I
would suspect that Solaris servers would be easier to administer remotely.
Would a larger tape library make more sense, if so what's available? Is there a
competing tape library that's more scalable? What type of Sun would be able to
handle at least an OC-12 worth of throughput? a Gigabit ethernet worth of
throughput? Would OC-12 be better or the Gigabit ethernet? (I'm thinking that
with the OC-12, it may be possible for the bp.start and end scripts to establish
a route with guaranteed bandwidth for each client thereby allowing the ATM to
allocate it's contracts more closely).
Regards,
John I Wang
Sr. Systems Engineer
Steverson Information Professionals
---
Enron Broadband Services
Enron Building 1472c
ph (713) 345-4291
fax (713) 646-8063
|
