On 7/19/15 7:13 PM, Troy Benjegerdes wrote:

On Sun, Jul 19, 2015 at 06:58:18PM -0700, Stephen Williams wrote:

..
If you are large and/or savvy enough, the thing to do is to borrow
cloud system methods and run a cloud for yourself.  Currently,
that's not completely easy or turnkey.  At some point, we should get
to a clean utility computing model, but it will take a few more
generations of evolution.

sdw


IBM would tell you the z13 is the best platform to run a cloud on.
Claims are you get 8000 or so cloud servers per machine [1]. I'm
sure fujitsu or some other vendor will sell you something equally
expensive in the same 'mainframe' class that can virtualize like
that.

A lot of what I hear about 'cloud' and virtualization are things that
were first deployed in 1970's-ish on mainframes.

Now, you're absolutely right that a 1TB hard drive that has been
qualified to work with that machine will cost about 10x what you can
get at staples.

   It's 10x for the drive, another 10x for the box to put it in, another
   10x for a license for the software to get to it, ...  (Roughly. ;-) )
   You can be nickle and dimed up front or over time.  In the latter case,
   it will continue to get more competitive and begin to have local
   systems with the same characteristics.
   How much does an additional 4TB of storage for a z13 cost?

But the point about mainframes is they are built to have lots of
*memory bandwidth*, and a 'compute minute' on a Z13 is going to process
a lot more transactions and write them reliably to that overpriced
disk than any cloud solution is ever going to do.

   Most cloud systems fall into the embarrassingly parallel category.
   Many smaller, cheaper, cooler units completely outclass, in price and
   scalability, bigger, faster, higher bandwidth solutions, unless those
   are built inexpensively with smaller, cheaper, cooler units.  We're
   finding out whether medium sized (Intel/AMD desktop / server class
   CPUs) or small (ARM mobile chipsets) are going to scale better, but
   either way, a many node system has an aggregate memory bandwidth that
   dwarfs old-style mega CPU systems.  It's not clear, but it appears that
   the z13 is just an integrated cloud-style clustered system with a bunch
   of nice added features[2].  If so, which is the only way it could
   compete on scale and cost, it is a branded cloud system.  Would it
   really be less expensive to operate than an Open Compute local cloud?
   Probably only if you made a lot of assumptions about overhead, etc.
   The z13 looks cool, and has a lot of interesting features.  It will be
   interesting to see how it does.

You just have to be ready to write a check for a couple of million
if you want one of these things on-site, and that's why the cloud
exists, for the folks that either don't have that kind of money, or
don't understand why they should spend it up-front, instead of getting
nickel and dimed to death by cloud vendors and their hackers.

   If you have the type of business where you know what you need and how
   much of it you need, you can competitively provision a local solution,
   although there are still plenty of ways to go wrong.  And many do,
   IMHO.  Many businesses have relatively modest needs, don't know what
   their growth will look like, etc.  Large up front costs are bad in a
   lot of situations, as is committing to a certain scale when there is a
   lot of uncertainty.
   The number of businesses and organizations who fit that narrow
   situation are few and dwindling.  Sales will be able to rope in plenty
   more for a while, but for many it is not a sane choice.  Security
   breaks are mostly about passwords, trojans, spear phishing, zombie
   machines, etc.  For every possible exploit of a cloud system, which at
   the infrastructure level should have well-funded security, I feel there
   are many more gaps in the typical local alternative: Sloppy, old
   Windows systems with a sloppy network, open to everyone file servers,
   poor access control, terrible custom programming, no significant
   physical security, etc.  The best systems + networks + policies +
   personnel are more secure, everyone else is just lucky not to be
   targeted.
   This covers some of this territory:
   [1]https://news.ycombinator.com/item?id=2482123


[1] [2]http://www.computerworld.com/article/2872096/ibm-s-z13-and-the-case-for-t
he-mainframe-cloud.html

   [2] [3]https://www-03.ibm.com/press/us/en/pressrelease/45808.wss
   sdw

References

   1. https://news.ycombinator.com/item?id=2482123
   2. http://www.computerworld.com/article/2872096/ibm-s-z13-and-the-case-for-the-mainframe-cloud.html
   3. https://www-03.ibm.com/press/us/en/pressrelease/45808.wss