On 6/9/2011 20:27, Darren New wrote:
> The main difference between mainframes and smaller systems is that
> mainframes are optimized for I/O. Even the mainframe I used 30 years ago

Another example: The mainframe knew which disks were fast (having multiple 
heads per platter, for example) and which were slower, and it put the 
indexes of databases and directories and such (etc) on the fast disk and the 
bulk data and actual file contents on the slow disks.  It also allocated 
random-access files on cylinder boundaries.

Nowadays, the blades and desktop machines don't even know what the cylinder 
boundaries of their disks are.

-- 
Darren New, San Diego CA, USA (PST)
   "Coding without comments is like
    driving without turn signals."

Post a reply to this message

Le 2011-06-09 23:27, Darren New a écrit :
> On 6/9/2011 18:34, Francois Labreque wrote:
>> There's also the question of performance. One needs a metric ton of
>> bladeservers to be able to match the performance under load of a 20
>> year old
>> mainframe sysplex.
>
> The main difference between mainframes and smaller systems is that
> mainframes are optimized for I/O. Even the mainframe I used 30 years ago
> could do several I/O operations simultaneously faster than the CPU could
> handle it. For example, you could be swapping in/out three processes, as
> well as accessing the data and an index page of a database, while the
> CPU is running full speed. There were four IOPs for each CPU, and each
> IOP could handle two DMA channels.

Yep, or having separate front-end processors dealing with keeping the 
user sessions alive, etc.  Leaving the CPUs deal with important stuff.

>
> Sort of like how video cards nowadays do all kinds of processing without
> the CPU's involvement other than setting them up.
>

Or some high-end server network cards implement the basic functionality 
of the IP stack, or SSL encryption in hardware.

-- 
/*Francois Labreque*/#local a=x+y;#local b=x+a;#local c=a+b;#macro P(F//
/*    flabreque    */L)polygon{5,F,F+z,L+z,L,F pigment{rgb 9}}#end union
/*        @        */{P(0,a)P(a,b)P(b,c)P(2*a,2*b)P(2*b,b+c)P(b+c,<2,3>)
/*   gmail.com     */}camera{orthographic location<6,1.25,-6>look_at a }

Post a reply to this message

Le 2011/06/10 00:39, Darren New a écrit :
> On 6/9/2011 20:27, Darren New wrote:
>> The main difference between mainframes and smaller systems is that
>> mainframes are optimized for I/O. Even the mainframe I used 30 years ago
>
> Another example: The mainframe knew which disks were fast (having
> multiple heads per platter, for example) and which were slower, and it
> put the indexes of databases and directories and such (etc) on the fast
> disk and the bulk data and actual file contents on the slow disks. It
> also allocated random-access files on cylinder boundaries.
>
> Nowadays, the blades and desktop machines don't even know what the
> cylinder boundaries of their disks are.
>

You effectively can't know where the boundaries are when the drive never 
tels you the truth about it. The drive's organisation have been 
virtualised a prety long time ago, and each true cylinders may not have 
the same number of sectors as it's neibors. The OS sees all cylinders as 
having the same capacity. It also see a sectors, cylinders and heads 
count that don't have anything to do with the actual values.


Alain

Post a reply to this message

On 6/10/2011 10:52, Alain wrote:
> It also see a sectors, cylinders and heads count that don't have
> anything to do with the actual values.

Yep. That's what I said. Of course the drive knows which sectors are where. 
THe interfaces might not, and the OS can't see it, and that's exactly one 
reason why people still use mainframes where important information like this 
isn't hidden from the OS.

-- 
Darren New, San Diego CA, USA (PST)
   "Coding without comments is like
    driving without turn signals."

Post a reply to this message

Darren New <dne### [at] sanrrcom> wrote:
> On 6/10/2011 10:52, Alain wrote:
> > It also see a sectors, cylinders and heads count that don't have
> > anything to do with the actual values.
>
> Yep. That's what I said. Of course the drive knows which sectors are where.
> THe interfaces might not, and the OS can't see it, and that's exactly one
> reason why people still use mainframes where important information like this
> isn't hidden from the OS.
>
> --
> Darren New, San Diego CA, USA (PST)
>    "Coding without comments is like
>     driving without turn signals."

That's not quite true anymore, at least with IBM mainframes. That was true with
the early IBM drives like the 3330, 3350, 3380, and 3390 drives.  But the last
several generations of IBM disk drives have been raid devices.  The legacy 3390
disk drive format is emulated by heavily cached controllers, backed by hi-speed
scsi drives.  What appears to the mainframe as a cylinder is really data blocks
spread across several scsi drives.  The model ds8100 I work with emulates about
1200 3390 virtual drives with a total capacity of 15TB.  The actual data is
contained on a hundred or so 10000 RPM scsi drives.  All the time I once spent
trying to optimize disk performance by placing system datasets at specific
locations on the drives to minimize head movement is no longer necessary.

Isaac.

Post a reply to this message

> Darren New<dne### [at] sanrrcom>  wrote:
>> On 6/10/2011 10:52, Alain wrote:
>>> It also see a sectors, cylinders and heads count that don't have
>>> anything to do with the actual values.
>>
>> Yep. That's what I said. Of course the drive knows which sectors are where.
>> THe interfaces might not, and the OS can't see it, and that's exactly one
>> reason why people still use mainframes where important information like this
>> isn't hidden from the OS.
>>
>> --
>> Darren New, San Diego CA, USA (PST)
>>     "Coding without comments is like
>>      driving without turn signals."
>
> That's not quite true anymore, at least with IBM mainframes. That was true with
> the early IBM drives like the 3330, 3350, 3380, and 3390 drives.  But the last
> several generations of IBM disk drives have been raid devices.  The legacy 3390
> disk drive format is emulated by heavily cached controllers, backed by hi-speed
> scsi drives.  What appears to the mainframe as a cylinder is really data blocks
> spread across several scsi drives.  The model ds8100 I work with emulates about
> 1200 3390 virtual drives with a total capacity of 15TB.  The actual data is
> contained on a hundred or so 10000 RPM scsi drives.  All the time I once spent
> trying to optimize disk performance by placing system datasets at specific
> locations on the drives to minimize head movement is no longer necessary.
>
> Isaac.
>

All this talk makes me wonder... has there ever been a mainframe port of 
POV-Ray?

/Sorry for veering on-topic
-- 
/*Francois Labreque*/#local a=x+y;#local b=x+a;#local c=a+b;#macro P(F//
/*    flabreque    */L)polygon{5,F,F+z,L+z,L,F pigment{rgb 9}}#end union
/*        @        */{P(0,a)P(a,b)P(b,c)P(2*a,2*b)P(2*b,b+c)P(b+c,<2,3>)
/*   gmail.com     */}camera{orthographic location<6,1.25,-6>look_at a }

Post a reply to this message

On 14/06/2011 02:04 PM, Francois Labreque wrote:

> All this talk makes me wonder... has there ever been a mainframe port of
> POV-Ray?

1. Yes, almost certainly.

2. Why would that be useful? Mainframes are supposedly optimised for I/O 
performance, not compute performance.

Post a reply to this message

> On 14/06/2011 02:04 PM, Francois Labreque wrote:
>
>> All this talk makes me wonder... has there ever been a mainframe port of
>> POV-Ray?
>
> 1. Yes, almost certainly.
>
> 2. Why would that be useful? Mainframes are supposedly optimised for I/O
> performance, not compute performance.

Ever heard of Cray Supercomputers?  ;-)

-- 
/*Francois Labreque*/#local a=x+y;#local b=x+a;#local c=a+b;#macro P(F//
/*    flabreque    */L)polygon{5,F,F+z,L+z,L,F pigment{rgb 9}}#end union
/*        @        */{P(0,a)P(a,b)P(b,c)P(2*a,2*b)P(2*b,b+c)P(b+c,<2,3>)
/*   gmail.com     */}camera{orthographic location<6,1.25,-6>look_at a }

Post a reply to this message

On 14/06/2011 02:42 PM, Francois Labreque wrote:

>> 2. Why would that be useful? Mainframes are supposedly optimised for I/O
>> performance, not compute performance.
>
> Ever heard of Cray Supercomputers? ;-)

Depending on who you ask, a supercomputer isn't the same thing as a 
mainframe.

When I was at the London Science Museum, I had a chance to see a Cray 2 
supercomputer. According to Wikipedia, this was the fastest computer on 
Earth when it was first built, managing to achieve a peak performance of 
about 1.9 GFLOPS.

...which sounds like a lot, until you realise that the Cray 2 originally 

US) which produces about 2,488 GFLOPS.

(Also: Trying to get GFLOPS ratings for recent CPUs is like trying to 
find gold dust!)

Post a reply to this message

Francois Labreque <fla### [at] videotronca> wrote:
>
> All this talk makes me wonder... has there ever been a mainframe port of
> POV-Ray?
>

I did port a of Povray 3.5 to Linux390, a version of linux that ran on the
IBM 390 series mainframe.  Actually it was very simple,  download
the povray linux source, and do the standard install.  I can't
give any time comparisons since the install was on a mainframe emulator,
not a real machine.  If and when I get approval, I would like to install
linux on our mainframe, and of course one of the first things I will
do is install povray.  I don't expect it to be that fast compared to
a standard pc.  Processor speed is not that much faster than todays
pc's. It has the standard ieee floating point instructions, but its optimized
for commercial business processing, not scientific or engineering type work.
Plus I'll have to come in at 3AM on a Sunday morning when the business workload
is at its lightest.

Isaac.

Post a reply to this message