Tracy R Reed said:
> Wade Curry wrote:
>> paragraphs :-) . The breach between the mainframe and *nix camps
>> is more cultural than anything else, IMHO. We tout our clusters and
>> distributed setups, and then install a blade server to get them all
>> in the same rack. They tout the blessings of consolidated
>> hardware, higher throughput and ability to get everything out of a
>> CP (they are designed to run at 100% capacity), and then install
>> dozens of virtual machines...
>
> I would definitely like to learn more about mainframes and apply the
> good ideas from the mainframe world to the Unix world. I understand that
> mainframe hardware is of the highest quality and fault tolerant with
> lots of emphasis put on IO and I hope we can carry more of that over
> into the Unix/PC world. I am still trying to understand what advantages
> the mainframe software world has. My biggest complaint (which I recently
> voiced on slashdot) is that we have no way of learning about mainframes.
> The average person can't afford one. The average (small) company can't
> even afford one! And Wade is only the second person I have ever met in
> my life who had any real experience with mainframes. I think mainframes
> would be a lot more widespread if IBM had given the software away for
> free and some minimal hardware platform on which to run it so that more
> people could have access to understand the technology. I know I will
> never recommend a mainframe solution to my employer until I really have
> some practical experience with one. And given the amount of OLTP and
> OLAP that we do around here we could probably use some of those
> capabilities of a mainframe.
Even the mainframers of the world have been yelling this at IBM. This last
announcement may not seem ground-breaking, but to a lot of folks it's looking a
lot
like a thaw has begun at IBM. If they start to look attractive to medium sized
businesses, the average unix guy may end up working with/around them a lot
more..
and to my mainframe newbie eyes, the technology does look like it will be
merging in
the future, too.
>
>> The *nix camp though, seems to be absorbing less of the better
>> technology from mainframes... and doing it more slowly. Xen seems
>> to be a good example. On the other hand, there's not a whole lot
>> else. We don't have nearly as robust facilities for scheduling
>> batch jobs, allocating resources to specific tasks on the fly,
>> detailed transaction logging, etc.
>
> What do you mean by "scheduling batch jobs"? I understand that "batch
> jobs" are a major part of the workload for mainframes but I have no idea
> what they really are. Would a batch job just be the generation of a
> report or something? If so we schedule batch jobs out of cron and it
> works just fine.
Sometimes it is something as simple as a report generation job, or an image
copy of
a DB2 database dataset. It really goes a lot further, though. Batch jobs are
used
for automating a vast number of things. Some online applications have a batch
stream of 2 thousand jobs in a single night! It is possible to have a batch
job run
as a temporary "message processing region" for an IMS application. (think of
IMS as
an application framework/server.. kinda) If you wanted to use builtin
functions of
an application that was written for use under IMS. You can write a batch job
that
attaches to IMS and automates actions that might otherwise require a human at a
terminal. Batch controls the availablity of the online applications
themselves, and
plays a part in automating the "checker" processes that looks for possible
problems
every few minutes of the day.
FWIW, there are several vendors selling batch scheduling software. It is able
to
direct jobs to the LPAR that has the correct environment, or the least load.
It is
aware of the sysplex, and can send jobs to other systems completely. It can be
made
aware of a variety of kinds of "triggers" to start jobs or batch streams,
whether
that is file activity, the clock, completion of a previous job or cycle...
there's a
lot to it. Check out ESP from Cybermation. (unfortunately Cybermation just got
bought by Computing Associates, who has a history of being the gorilla that
kills
good sw from smaller companies.)
The unix world would get less out of ESP than mainframes, but it might benefit
some
shops to have a lot more control over their batch processing. ESP does run on
Linux, I'm told.
> Maybe the mainframes have a better way of doing it but
> the way we are doing it is "good enough". I would like to be able to
> allocate resources to specific tasks on the fly but I think we are
> getting there with clustering and Xen. Detailed transaction logging is a
> part of our application. Is it part of the OS in the mainframe world? If
> so, why should the OS need to be aware of Alice buying a toothbrush?
>
This is not easy to figure out. The way that you approach the workload is
influenced by your tools sometimes. Mainframes have more logging than I think
they
need. On the other hand, it has to deal with so much volume, that it is very
good.
And, if you really, absolutely, can't lose a single transaction... well, you'd
better have them in a log to "replay" and bring databases up to date. The
volume in
a case like that is huge, but considered unavoidable here in my shop. Of course
they mitigate the huge storage requirements by using tape libraries. The OS can
have a facility installed that will automatically age and migrate datasets off
of
disk and onto tape in stages.
I really don't want to take the approach here of saying one methodology is
better
than the other. At least not at this point. But if you put all the big iron
aside,
you still end up with a mindset that is more geared to what mission critical
business computing requires. While both camps may debate the power and
reliability
of OS and HW, I think the suits just sit back and say, "who gives a rodential
hindquarters about 5 or 7 nines? I get to control how much CPU is used for a
single
application! Control! HAHA!" I have a jaded view of upper management. ;)
In any case, because of IBM and the general computing environment, mainframers
are
often having to implement Linux, and/or some sort of distributed software, p2p,
thingamajig. Most of them hate it. But their learning. The unix side is
handicapped, though, because we don't get to see the mainframe side as you
mentioned
to begin with.
>> The blog link mentioned below is a fairly good one for mainframers.
>
> I will definitely be checking it out.
>
>> In case you need a reminder, IFLs are specialized CPUs that
>> are optimized for Linux.
>
> If you run Linux on a mainframe don't you lose the batch processing and
> other niceties that the mainframe OS provides you? Then what you have is
> a very reliable Linux box on very expensive hardware. Still better than
> running it on a PC, I'm sure.
>
Not necessarily. Mainframe HW is able to run multiple LPARs (Logical
PARtitions)
regardless of the OS(s) you install on it. You may want extra security for one
instance of Linux for firewall or other purposes. So you allocate enough
resources
for that instance of Linux and install other OSs on the others. I only know my
own
shop, but I've never seen MVS and VM together. VM is made to be able to host
other
OSs, even MVS, and increase the number of virtual machines that you can run.
(which
is why Linux is usually run under VM). At that point you get the benefits of
mainframes and Linux. The community seems to be divided, though. VM is either
wonderful, or is unstable and experimental. Depends on who you ask.
None of the mainframes I have access to have VM or Linux, unfortunately. :(
Wade Curry
o _ _ _
_o /\_ _ \\o (_)\__/o (_)
_< \_ _>(_) (_)/<_ \_| \ _|/'\/
(_)>(_) (_) (_) (_) (_)' _\o_
--
[email protected]
http://www.kernel-panic.org/cgi-bin/mailman/listinfo/kplug-list
