Re-hosting production application on Open Systems

, , , ,
1

Under zOS, my client processes 1 billion Adabas calls per week against 350G of data. Several development and test databases have proved that the components of the primary Natural/Adabas application will run on a Windows server. Now we are considering re-hosting the production application to Open Systems (Natural for Windows, and Adabas for Windows or Unix/Linux), but we have many questions. Here are a few:

1 Has anyone run this size of (production) load (1G calls; 350G data) on a Windows or Unix/Linux box?

2 Does Adabas take advantage of Windows or Unix/Linux features that benefit performance? If so, are the benefits so great as to simplify our platform selection?

3 What were the compelling reasons for your choice of operating system?

Thank you.

2

Hi Ralph;

I have noticed that Adabas (and Natural) performance is greatly improved as memory is made available (more so than on mainframe).

I went looking to confirm that Natural still has a limitation of 8 records for MultiFetch. This seems to be absent from the latest documentation. So, either the limitation no longer exists, or the documentation suffered in the last re-write. will check tomorrow.

steve

3

55G data, ~ 100M calls, ADABAS 5.1.7 @ Solaris 10

My company did the migration before I stared to work there. But my collegues say that there was a big performance improvent compared to the old BS2000 system.

4

Steve,

I don’t have Nat637 yet, but 636 returns 8 records via Multi-fetch for a local database.

5

Hi Ralph;

Thanks; saved me the trouble of checking (i am also not on 6.3.7).

Clearly, if your application uses MultiFetch, this could be a significant difference if you were to pass a large file with say a MultiFetch of 50 or 100. The PC would then incur many more Adabas calls.

I will check the Natural documentation again. I know that there used to be a discussion of the 8 record limitation somewhere. I think I have an older Natural still on my laptop; should be able to find out where the discussion is there; then check the latest documentation.

steve

6

A discussion of Multi-fetch is too detailed for what I was asking in the original post. We know that every batch and on-line component of our application will run reasonably well under Open Systems (we tested with Windows 2003), and we’ll run benchmarks to determine how big a box is needed to approach the throughput of our zOS system, but we’d like to know what other users have experienced, with our levels of volume and with the Windows/Unix/Linux selection.