Infra is not really a concern for me, it's more will the software handle it.
Network connectivity and IOPs aren't a huge concern (machines will be speced to be "beasts" as well as network.
de-dupe for me is not a massive want. 99.9% of the files we will be backing up are 100% unique. I'd rather spend the time shuffling data then de-dupe it.
For me it's more of a is BPC even a remotely viable solution. Other studio uses netbackup and I'd rather sit on a razor than use it to be honest so I'm out fact finding.
I've used BPC v3 but to back up /etc off servers. Never to handle PBs of data.
It would be interesting to hear if anyone is using it at that scale...
I expect that if you were going to try that, then you should use BPC v4
(because it has the ability to avoid transferring a file if another
client has already backed it up) and assuming you are going to use rsync
everywhere.
I would also think that you will need to use multiple servers, otherwise
you will spend a huge amount on getting enough CPU performance in the
machine (all that encryption, compression, etc takes CPU), and RAM
(storing lots of data and cache in RAM) and also the massive IOPS
required. Consider the number of clients, you say 100s, so assume 300.
You need a backup every day, and assume each backup takes 45 minutes.
You need a concurrency of at least 30 to complete all the backups in
under 8 hours (assuming an 8 hour backup window). OTOH, you could have 6
servers each handling 5 concurrent backups, and hence have 50 clients on
each server. This loses some of the de-dupe, but it makes everything a
lot easier.
Personally, you could extend this further (depending on budget, risk,
etc) and have 12 servers, such that each client is backed up by two
different servers. A failure of one server or its underlying storage/etc
will mean you do not lose all the backups.
Someone (on the list earlier) suggested that it would be possible to
have BPC v4 use an object storage backend, so as long as that backend
could perform fast enough, then it may be one way to get a larger "farm"
of BPC servers but still keep de-dupe across all of them.
------------------------------------------------------------------------------
Check out the vibrant tech community on one of the world's most
engaging tech sites, SlashDot.org! http://sdm.link/slashdot
