Now I would be asking the opposite question, how many tape drives per media server, based on the guidance I’ve seen somewhere about how many MHz of CPU are required to drive a tape, and how many MHz are required to handle each Ethernet port over which the data arrives.  There is a lot of wet finger as there is always a bottleneck somewhere, it is just very time consuming to chase them down.

Our older sites use SSO a lot and it is a bad thing as it was overused – servers wait many many hours for drives to come free, and can even wait mid-backup when they change tapes; NetBackup not unreasonably tries not to use the drive that is unloading for the next tape, but that means that a server changing tapes just joins the queue at the back.

Newer site there is no SSO; we have Sun T5220 media servers each with 4 x LTO4, 2 on each of 2 x 4Gb/s fibres.  That is a theoretical bottleneck because if both drives are doing 2:1 compression we are over what a 4Gb/s fibre can do.  We use disk staging to prevent the slow clients hogging drives, so all tape access is SLP duplications which run at at least 100MB/s.  The media servers are also FT media, so what would have been SAN Media Servers (big RAC systems) are SAN clients.  Much easier to manage, seems to run OK.  Hopefully no shoe-shining.

Doesn’t answer your question at all!  If you really do want to use SSO then it comes down to seeing how many hours each media server needs a drive, seeing if you can create backup windows that are as wide as possible, and fitting the hours you need a drive into the hours in the backup window.  Go above that and Status 196 will hurt you.