How much disc for a 2TB database?

[MikeRBS]

I have an archive application that is about 2 terrabytes of data and will run on a big mainframe. It will hold data about old insurance policies (several million) for maybe 10 years. Initially there might be about 400 enquiries per day, falling away to maybe 1 enquiry per day after say 5 years. Response time can of course can be degraded compared to a live application, and can slow down as time progresses.

What is the minimum amount of real disc that needs to be added to the mainframe environment concerned? Perhaps a more realistically, what sorts of things would you factor into this assessment?

 

[IDC3044]

We calculate disk space as 48K per track or 720K per cylinder. We have Mod9 DASD that hold 150,255 tracks or 7.21224 Gigabytes. Using hardware compression you may be able to squeeze around 21 Gig onto one disk drive.

 

[MikeRBS]

I was thinking in terms of a heirarchical storage environment. With a large application presumably at least some partitions could be in mass storage (automated tape or whatever people use nowadays). Also at any time some pages I guess will rolled-back to near-line (or whatever slower disc is called now) if they haven't been accessed for a few hours.

When you get to an archive scenario this effect grows. In the extreme case where no-one accesses the data for months the entire database might have been moved to backing storage. Realistically there will be something in between with most of the data off front-line discs but pages and partitions associated with data in use on near-line and main disc packs.

Obviously this depends on the database design and usage level. I thought however someone might have seen this type of low usage-level/large database and could say "We managed to run System X with only 10% of the data on main disc" Or any information on constraints eg maybe you have to have at least one partition of all tables on disc.

In my situation front-line disc is very expensive because it has to be mirrored, DR-ed etc due to intransigence of operations personnel.

 

[stevexff]

Have you considered using a data compression exit? You would be trading CPU for disk space, but if you only expect to service 400 requests a day, it would be a bargain.

Also, is the mainframe the right choice of platform? You are serving 400 requests a day here, not 400 per second. A couple of PC servers (for failover) and a RAID would do the job. Multiply by two for DR, and it's still way cheaper than the mirrored DASD...

 

[MikeRBS]

I think data compression would reduce the space taken up by the data by 60-80% which is useful but I suspect a strategy to keep as much data off disc as possible would yield much higher figures.

You are right - the choice of a mainframe is entirely wrong. We're not even re-using existing code. It's just political but that happens a lot in big corporations. You have to live with it.