[REQ] Help with sizing a machine to feed a tape drive

I am thinking of buying a tape loader [probably a Tandberg StorageLoaderâ„¢ LTO-5]

Now I would need a machine from which the tape loader could be run, some form of machine to collect data and then write it to tape. Call this machine CONTROLLER

My idea would be to collect the data which is to be written to tape on the CONTROLLER. Data would be collected from a number of sources, such as slow NASes. When the collecting is finished, data would be dumped to tape.

How should such a CONTROLLER be dimensioned? As far as I was able to determine it needs to be able to provide a sustained input for the tape drive: ie > 0.5TB/h or > ca 140 MB/s. I would thus need CPU, RAM and an HDD which can provide this. Would something like a Intel Xeon E2-1230 with 4 (or 8) GB of RAM and a RAID-1 of two standard drive (say WD RE4 with sustained read specified at 110MB/s) be sufficient, or would more impressive hardware be required? - What would be necessary here: more spindles or something else?

Is this the correct way to arrange for writing data to tape? I would probably not have much data which changes - so relatively little local storage would be necessary. Probably no more than say 3 TB (ie one SATA spindle).

Price isn't the main factor at the moment, as this is more a fact-finding exercise than anything else. I'm primarily looking for other people's experiences and best-practices than any actual hardware recommendations.
 
Will an RE4 drive really deliver that 110M/sec? I suspect it's a best-case number.

Mirroring (RAID1) might speed up reads. gmirror(8) has several balancing algorithms; benchmarking is the only way to really know if any of them are quicker in reality.

Other RAID configs really can go faster. CPU and memory probably aren't too critical for 140M/sec.
 
wblock@ said:
Will an RE4 drive really deliver that 110M/sec? I suspect it's a best-case number.
It seems to be a best-case number. But for example the Constellation ES.2 drives are speced at 155 MB/s. I might just have to buy more expensive drives.

Other RAID configs really can go faster. CPU and memory probably aren't too critical for 140M/sec.
Which one should one chose for the highest read speed? RAID10? For discussion purposes we might assume that there are four drives.
 
Various striped configurations are faster. Whether the risk of RAID0 or the complexity of RAID0+1 or RAID1+0 is worth it...

At this point, asking on the freebsd-questions mailing list would probably get a better representation of what people have actually used and what throughput they got. My experience is too limited.
 
Back
Top