Thursday May 22, 2008
Thursday May 22, 2008
The marketplace and hollywood have very restrictive views and impressions of the mainframe and many believe that IBM are the only ones that can supply such a thing. Well in reality the mainframe does not exist only in the minds of marketing and fiction. As many suppliers nowadays have equivalent and/or superior mainframe systems to IBM's.
Everyone who has not worked with mainframes has this impression; they are large, expensive, fast, complex, special, adaptable, reliable.
Everyone who has worked with mainframes knows the following; they are large, expensive but will be discounted on the hardware to make them look competitive, but software charges and maintenance will be expensive, they are not fast anymore or no faster than any other large system, you can only run zOS (MVS), they are not complex just totally proprietary. Not really adaptable, everything needs to run on the zOS instruction set need hypervisors (read overhead) to do conversions to run other interesting OSes like Linux, really good for the old stuff like CICS and old batch programs that no-one understands as the people who wrote the retired. They are special as a whole load of emotional baggage goes with them, companies cannot migrate from them as no-one understand the apps, so companies are held hostage, costs are often hidden in various ways (s/w, maint, service) so you can never get to the bottom of the TCO. As for reliability no different than any other major large system where you have good change control. Mainframes have good reliability records as no-one is allowed to change them. If you have other high end systems that you do not change they will be as reliable as a "mainframe". The mainframe hardware is not special compared to other high end systems. Historically they were better about 10yrs ago, but this myth continues.
So we either stop calling the z10 server a mainframe or we call every high-end server a mainframe. Sun and IBM high end Unix mainframes (servers), (not sure about HP, SGI etc) high end servers have better or equivalent features to the z10 servers.
Mainframe acid test, ask IBM what is the best plaform for every application that you want to run, what is the difference between their Linux actually Red Hat, AIX and zOS offerings. Is AIX inferior to zOS (aka MVS), is Red Hat not as good as AIX. Is Red had better than zOS. When should we use what.
Can I run Red Hat on a p-series
Can I run AIX on a z10 mainframe server (apparently it uses the same CPUs)
Can I run zOS on a p-series.
I get really confused with the positioning and confusion of AIX, Red Hat and z/OS. When do I do what with what. Or do I just do everything with everything and make a real mess. Good fun for the techies.
So with HP recently they came up with some logical partitioning with something called Dynamic LPARs, well that is old established technology, nothing new here. Available from others and established technology for a long time. However, HP engineering was always good, HP used to be a engineering company initially. So their high end UNIX servers can also be looked upon as mainframes.
Now Sun M-series servers have instruction retry, whole memory DIMM mirroring, crossbar ECC, more dynamic replacement of CPU's, RAM than any other supplier while the system is running and more. So you not get get more reliable than that and mainframes may not even have the reliability of some high end Unix servers.
Mainframe has been around since 1960's so the name is familiar but missused. Like many things in life, understanding, education and knowledge helps dispell these myths when once upon a time long ago they were something special. Now I like the mainframe, love to tune the assembler code, patch and zap modules, good for your old apps that no-one understands and your company is too scared to migrate to a new lower cost system. But as with everything technology moves forward and other high end systems especially large Unix servers are just as good or better and can be understood by more people.
It does seem that OS's are becoming increasingly interoperable with hardware platforms that differ from the original kit vendors. e.g. we're seeing this with RHEL and SuSE on ZoS, p-series and Itanium. Also Solaris x86 on a multitude of x86 platforms. This is also creating revenue oportunities for service & support. Solaris on ZoS has been rumoured but nothing solid seems to have been confirmed so far.
Seems that IT shops who choose to write their own application code can always target their most effective platform combination for a given workload. Whilst if you're into buying 3rd party packages then 9/10 times you need to pay special attention to what the vendor is developing on. Otherwise you risk ending up with something really niche with poorer vendor support.
Posted by Tom on June 16, 2008 at 01:10 AM CEST #
Cannot argue with that, it is getting better for customers when they have binary compatibility between OS on different platforms and the look and feel is the same. Things are a lot better nowadays. As you can move from a X64 to large RISC server without having to change your OS or learn anything new. Just get more RAS and scalability as you go. We have to thank Unix and open systems for that, lots of suppliers wanted tie in to a specific CPU and OS. Now the successful OSes run on different CPU's.
Posted by valdis on June 18, 2008 at 11:28 PM CEST #