>I would like to help you. But my experience with RAID is rather limited. This is second hand information.
>
>Redundent Array of Independent Disks.
>

There are lots of levels of RAID functionality. RAID 1, mirroring, writes thes ame data to multiple physical targets simultaneously, and may allow for split seeks. RAID 0 is striping, writing part of a datablock across multiple target devices, extending avaialble media bandwidthat some cost in latency with no enhancement in reliablitiy (in fact reducing reliability because the number of physical mechanical devices is in creased, and the failure of one results in data loss). RAID 3/5 introduce the use of a parity mechanism that can be used to recover from the failure of a single physical device by recording parity information (3) on one physical device or (5) spread across all physical devices at the net sacrifice of 1/n (n being the number of devices in the array) fraction of the total storage of the sum of the array storage to data recovery.

Duplexing introduces multiple redundant disk channels, controllers and power supplies into the mix for increased reliability. Truely mission-critical data can be maintained in duplexed RAID-5 arrays (each RAID-5 array is mirrored to another identical RAID-5 array in parallel with lots of redundant hardware.)

RAID is discussed in detail in any number of good whitepapers; some of the clearest were written by DPT (now a part of Adaptec), in part reprinted in Peter Ridge's "The Book of SCSI" from No Starch Press (ISBN 1-886411-02-6). I believe the DPT whitepapers may still be available through Adaptec's web site. From an NT/Win2K perpsective, there are discusssions of mirroring/duplexing and the striping issues in the MSDN Library docs, part of the NT Server and Win2K Server docs in particular. There are more technically in-depth examinations of RAID and alternative technologies like PTD in other books.

If real issues of availability and reliability are at stake, investing in a knowledgable professional is needed; the reading ain't a matter of an email message, and there are literally books devoted to the topic of RAID alone. The worst mistake to make is to try to bs your way through the topic.

>I will tell you what we do when we need to make a backup when people have work to do. We kick everyone out of the system. We copy all the data to a temporary directory. We allow everyone back into the system and we run the backup on the copy of the data in the temporary directory.
>

That works for static snapshotting; the reliability of your journalled transactions from the point of snapshotting determines the amount of data at risk from the time the snapshot is taken. This is the basis of checkpointing a process to protect the data at a known point in time.

>So we do have to kick people out for a period, but it is a shorter period of time than it takes to do the tape backup because we are copying hard drive to hard drive. This may not be a viable solution in your situation though.
>

There are plenty of storage solutions available that can move data to tape at least as quickly as data can be written to disk; DLT and AIT are two tape technologies that have data rates that can exceed the sustainable data rate off a hard drive. There are others as well.

My approach at Weatherhill as well as my clients' sites has been to snapshot the entire schema nightly during off-hours, and to snapshot transaction journals at periodic intervals during the working day, which reduces system unavailablity during our operational hours significantly and protects our committed activity with greater frequency. We also rigorously plan and implement checkpoint/restart processing on irreversable/uninterruptable processes.>