Mainframe vs Open Systems

steve6 · New User Joined: 10 Jul 2007 Posts: 7 Location: Chennai

Hi,

In my recent project which is a conversion from Mainframe to open systems, I came across an interesting question from a non-mainframe colleague. I guess I have answered that question, though not 100% accurate

- the immediate impact of my answer being minimal.

I thought I should evaluate my understanding and correct myself, if requied. Pls share your thoughts.

Here is the question from my non-MF colleague:
How come Mainframe handles around 1000 transactions per sec where as with open systems we can assure only about 10 transactions per sec?

Here is my answer:
1. Going into the granular level of transaction, our CICS programs use VSAM files which are open out there to process things faster
- no opening / closing required
- KSDS, AIX/Path play role in quick access

2. When we design the Mainframe Application, we see to that the VSAM files used by CICS are pretty much like carriers of data - not permanent storage.
- number of records being less, the processing would be faster

3. We use something called psuedo batch(that doesn't involve screens) that does the posting of data from the vsam files into the DB2 tables (as permanent storage)

4. Even when we use DB2, the parallelism (processing and I/O) helps.

(Its a large scale application)
(detail on the transactions... when we say 1000 transactions, those are 1000 records coming in through MQ and the processing response is sent back thru MQ)

Robert Sample · Posted: Tue Sep 28, 2010 4:09 pm

Mainframes are optimized for high-speed, high-volume data processing. I/O is handled by a separate subsystem, not by the CPU -- as one example. There are also specialized processors to handle specialized tasks, instruction look ahead, and many other features designed to support volume processing -- making a mainframe more like a cluster machine than a single box. Furthermore, the mainframe has benefited from the more than 40 years of development efforts aimed at improving processing throughput for large volumes of data.

steve6 · New User Joined: 10 Jul 2007 Posts: 7 Location: Chennai

Just wanted to mention couple more things.

1. I was just involved on the conversion part of the application from Mainframe end... so even I was surprised to see such difference 1000 to 10. I thought it would be the poor design in open system. but still the difference was extreme.

2. That project got over, the converted application went into shelves and just considered as a prototype in open system.

The application remains in Mainframe.

dbzTHEdinosauer · Posted: Tue Sep 28, 2010 4:25 pm

mainframe.typepad.com/blog/2010/09/the-worlds-fastest-microprocessor.html

Ronald Burr · Posted: Tue Sep 28, 2010 8:12 pm

Computers set for Quantum Leap

Will this new chip design make it possible for a computer to pre-fetch/pre-staticize MULTIPLE potential instruction sequences simultaneously - meaning absolutely NO interruption in processing speed for heavily branch-bound algorithms ( e.g. binary searches )?

dick scherrer · Posted: Tue Sep 28, 2010 11:46 pm

Hello,

One needs to consider that the mainframe has been "tuned" for heavy multiple-user activity. From the early days, the focus of the mainframe has been to support many unlike activities at the same time - and provide good thruput.

If by "open systems" you mean Windows, keep in mind that Windows is a Personal Computer operating system. Which has been re-worked to run on a server so multiple other Windows PCs can interact with the server(s). The Windows Server environment is still a long way from an industrial strength multi-user operating system. IMHO

If you mean Unix, several of the systems i've helped migrate from either the Mainframe or Windows run very well on the Unix systems. . . fwiw.

mtaylor · Posted: Fri Oct 01, 2010 5:28 am

The fact that DB2 and the application run on the same hardware is probably a big performance plus for the mainframe, since in most 'modern' technology, the database runs on a separate server (or cluster of servers) requiring network communication which can be thousands of time slower than the communication buses inside a box.

But this is quiet a nebulous topic; there's no way to say why your application was 100x slower without doing some indepth profiling and load testing, which the open system devs should've been able to do. There are simply too many variables involved. Ie it could've been poorly coded. Which is all to easy to do.

dbzTHEdinosauer · Posted: Fri Oct 01, 2010 5:55 am

Not that I am a fan of SAP, but SAP R3 can run 250-300 trans / second on a midrange server.

and a SAP transaction probably requires a minimum of 10-20 db2 table accesses.

now, the boys at SAP are no saps (pun intended). They know what they are doing.
which leads me to agree with mtaylor -
the people who wrote the open systems system did not.

dbzTHEdinosauer · Posted: Fri Oct 01, 2010 6:19 am

and the 250-300/sec was a benchmark from a project that I worked on in 2001.

now nearly 10 years later, I have no idea what kind of thru-put they have.

steve6 · New User Joined: 10 Jul 2007 Posts: 7 Location: Chennai

Thanks for the responses. I could see there are multiple aspects contributing to the Mainframe's capacity. It was really helpful for me to see what lies behind our green screens.

Anuj Dhawan · Posted: Thu Oct 14, 2010 4:55 pm

During 2004 I heard of that in US, they don't open a new firm until they promise to use SAP/ERP in their Firm! (though I've never veirfied this statement!)

dick scherrer · Posted: Thu Oct 14, 2010 8:20 pm

Hi Anuj,

Who are "they"?

I know of quite a few "new firms" that do not use SAP. . .

There is definitely something here i misunderstand

d

Anuj Dhawan · Posted: Thu Oct 14, 2010 8:27 pm