Anyone going to Supercomputers '09 in Portland? [Computer Architecture]

Prev: Performance of SMT on Atom
Next: Lipitor generic cheapest. Buy cheap Lipitor online. Buy Lipitor cod. Lipitor overnight delivery.

From: Anne & Lynn Wheeler on 15 Nov 2009 11:16

"Del Cecchi" <delcecchiofthenorth(a)gmail.com> writes:
> As I recall, most of the toolsrun thing was master slave, with a
> master usually in YKT and the shadows (slaves) remote, along with
> references to the scifi "7 princes in amber" or something like that.
> Reading was done from the local shadow. posting went to the master
> and was reflected back to the shadow.

fiber optic technology had been knocking around in POK since late
70s. one of the austin engineers took the technology, tweaked it, much
cheaper drivers ... and it was part of the RS6000 product as "SLA"
(serial link adaptor ... similar, faster, cheaper, but incompatible with
the POK mainframe ESCON).

chips were done by rochester. in an attempt to get SLA more acceptable,
we talked a router vendor into adding SLA support. Then we had to talk
Rochester into supplying the chips to the outside vendor. Rochester
would an inter-plant "transfer" the chips to Austin ... at 300% markup
.... and then Austin would "transfer" the chips to the vendor at 300%
markup ... total 900% markup ... for a vendor that was doing us a favor.

I had been doing various stuff on & off over the yrs with LLNL ... which
were driving force behind FCS standard. When the SLA engineer wanted to
start work on an 800-mbit version ... spent something like six months
convincing him to participate in FCS instead. He eventually did and
became the "owner" (secretary) of the FCS standards document.

Rochester and POK also started to participate in FCS standards (POK
channel engineers working hard on layering a half-duplex protocol on top
of the basic full-duplex FCS operation ... current FICON). There was a
standards FCS discussion list (fiber-channel-ext(a)think.com) ... but
Rochester also hosted an internal toolsrun discussion list
(dfcforum(a)rfcvmv) .. which included forwarding the
fiber-channel-ext(a)think.com traffic (as well as some other items like
the hippi discussion, hippi-ext(a)think.com).

misc. other stuff found its way on to dfcforum ... from long ago and far
away

MARKETPLACE NEWS

1. HP signed a letter of intent to buy Texas Instrument's multiuser
computer business with the intent to gain commercial market share
for its HP 9000 Series 800 computers. HP will encourage users to
migrate from the 125,000 installed TI machines to the 800 Series
servers. HP also acquired a well-developed reseller and integration
channel with an intimate knowledge of the TI user base.
Source: Systems & Network Integration Date: June 29, 1992

2. Stratus also announced that they would use HP's PA-RISC architecture
in future systems to be developed. Their director of systems products
explained that this was done after careful consideration of several
vendors architectures that would be available in the 1994-1995
timeframe. This is something of a coup for HP considering that 18%
of the Stratus 1991 revenue of $448 million was sold by IBM's
reselling of Stratus fault tolerent systems. As part of this deal,
Sratus will also port Unix 5.4 to the HP-RISC architecture.
Source: Systems & Network Integration Date: June 29, 1992

3. IBM will enhance its RS/6000 clusters this year by providing optical
channels between systems. Ancor Communications will provide the
optical communications between machines that can be located up to
2 kilometers apart. Phil Hester, AWD Vice President, said that this
technology will be Beta tested by year's end. "Loosely-coupled
RS/6000s have the ability to scale well beyond the power of ES9000
mainframes" said Stu Skomra, vice president of marketing at ILAN
Inc., a network integrator that uses RS/6000s, "but the downside to
this is that there is no single system image for systems administration."
IBM has yet to detail a strategy that allows clusters to be managed
administratively by a single image.
Source: System & Network Integration Date: June 29, 1992 Page: 12

4. Solaris 2.0 is experiencing performance impacts of 10-12% below
that achieved by Version 1.0 because of all of the code put in
it to support things such as multiprocessing according to users
that have tested the new software. Although 2.0 will not officially
ship until December of this year, the performance problems are
a major issue with integrators that are trying to use the software.
"Performance has been Sun's Achilles' heel" said Ira Cohen, president
of Copley Systems Corp., a network integrator. Solaris 2.0 is
a 32-bit Unix Operating System designed to support multiprocessing
and have hooks for OSF's Distributed Computing Environment.
Source: System & Network Integration Date: June 29, 1992 Page: 1

.... snip ...

unrelated topic drift ... recent mention of single-system-image
http://www.garlic.com/~lynn/2009p.html#43 From The Annals of Release No Software Before Its Time
http://www.garlic.com/~lynn/2009p.html#46 From The Annals of Release No Software Before Its Time

--
40+yrs virtualization experience (since Jan68), online at home since Mar1970

From: Del Cecchi on 15 Nov 2009 22:44

"Anne & Lynn Wheeler" <lynn(a)garlic.com> wrote in message
news:m3r5rzn4ql.fsf(a)garlic.com...
> "Del Cecchi" <delcecchiofthenorth(a)gmail.com> writes:
>> As I recall, most of the toolsrun thing was master slave, with a
>> master usually in YKT and the shadows (slaves) remote, along with
>> references to the scifi "7 princes in amber" or something like
>> that.
>> Reading was done from the local shadow. posting went to the master
>> and was reflected back to the shadow.
>
> fiber optic technology had been knocking around in POK since late
> 70s. one of the austin engineers took the technology, tweaked it,
> much
> cheaper drivers ... and it was part of the RS6000 product as "SLA"
> (serial link adaptor ... similar, faster, cheaper, but incompatible
> with
> the POK mainframe ESCON).
>
> chips were done by rochester. in an attempt to get SLA more
> acceptable,
> we talked a router vendor into adding SLA support. Then we had to
> talk
> Rochester into supplying the chips to the outside vendor. Rochester
> would an inter-plant "transfer" the chips to Austin ... at 300%
> markup
> ... and then Austin would "transfer" the chips to the vendor at 300%
> markup ... total 900% markup ... for a vendor that was doing us a
> favor.

That isn't how I remember it. I think the idea of using CD lasers
instead of the LEDs that POK used came from Rochester. The guys that
did it worked down the hall from me. I do remember the bizarre
contortions for Rochester to sell outside. I don't remember Austin
being involved unless it was Austin Research Lab. The group in
Rochester got moved to the Semiconductor Div for that very reason.
Then they got sold to JDS Uniphase which later shut them down so many
are back at IBM in Rochester.

I was told the FC-AL and phy came straight off the tundra. Rochester
was always doing own thing, with twinax links on 36/38 and FC-AL which
was like the twinax stuff only with optics instead of SSA and escon.
Fortunately Rochester was cold and remote so the corporate dweebs left
us alone.

>
> I had been doing various stuff on & off over the yrs with LLNL ...
> which
> were driving force behind FCS standard. When the SLA engineer wanted
> to
> start work on an 800-mbit version ... spent something like six
> months
> convincing him to participate in FCS instead. He eventually did and
> became the "owner" (secretary) of the FCS standards document.
>
> Rochester and POK also started to participate in FCS standards (POK
> channel engineers working hard on layering a half-duplex protocol on
> top
> of the basic full-duplex FCS operation ... current FICON). There was
> a
> standards FCS discussion list (fiber-channel-ext(a)think.com) ... but
> Rochester also hosted an internal toolsrun discussion list
> (dfcforum(a)rfcvmv) .. which included forwarding the
> fiber-channel-ext(a)think.com traffic (as well as some other items
> like
> the hippi discussion, hippi-ext(a)think.com).
>
snip

del

From: Anne & Lynn Wheeler on 16 Nov 2009 00:50

"Del Cecchi" <delcecchiofthenorth(a)gmail.com> writes:
> That isn't how I remember it. I think the idea of using CD lasers
> instead of the LEDs that POK used came from Rochester. The guys that
> did it worked down the hall from me. I do remember the bizarre
> contortions for Rochester to sell outside. I don't remember Austin
> being involved unless it was Austin Research Lab. The group in
> Rochester got moved to the Semiconductor Div for that very reason.
> Then they got sold to JDS Uniphase which later shut them down so many
> are back at IBM in Rochester.

re:
http://www.garlic.com/~lynn/2009p.html#85 Anyone going to Supercomputers '09 in Portland?

my epiphony to use cdrom components came on trip (mentioned here) to
other side of the pacific ... which was also first time to see consumer
electronics manufacturing using surface mount technology ....
http://www.garlic.com/~lynn/2009p.html#59 MasPar compiler and simulator

and came back pestering to use cdrom components (I don't know if that
predated rochester or not) .... i.e. $300 cdrom player was enormously
better technology than the 20times that I was paying for T1 modems
in HSDT project.

it predated meetings with LLNL and were pushing moving thier (ancor)
copper to fiber which evolved into FCS standards. this old post mentions
LLNL making claims about projected price/FCS-drop in '88 meetings.

Part of HSDT involved working with cyclotomics ... later bought by kodak
(in part because cyclotomics was responsible for lot of the FEC stuff
used in cdrom standard):
http://www.garlic.com/~lynn/2009e.html#66 Architectural Diversity
http://www.garlic.com/~lynn/2009n.html#46 Follow up

as per above reference ... some amount of it was reed-solomon ... and
HSDT project had fortune to have engineer that had been reed's graduate
student at caltech (and did some of the work on reed-solomon ... but
would claim that one of his favorite classes was undergraduate at MIT
from Anne's father).

In the early to mid-80s ... HSDT was involved in (telco) fiber links,
copper links, microwave links and satellite links. HSDT had dedicated
transponder in SBS4 and some of our own TDMA earth stations.
http://www.garlic.com/~lynn/subnetwork.html#hsdt

we had a TDMA earth stations built to our spec by a couple different
vendors. Somewhere along the way ... one of the vendors informed us that
a large national telco had approached them and asked if they (our
vendor) would build them (large national telco), a set of stations to
the same (HSDT) specs.

random drift from long ago and far away

Embassy of Japan
2520 Massachusetts Avenue
Washington, D.C.
10
To Whom It May Concern:

This letter is to advise you that the following individuals are
employees of the International Bussiness Machines Corporation,
and will be visiting Japan for one week starting Febuary 18, 1985.
Lynn H. Wheeler
xxxx
yyyy
zzzz

The purpose of this visit is to meet representatives of Matsushita
Electric corporation and to tour their facilities in Osaka, Gifu
Nagoya, and Tokyo.

International Business Machines will be responsible for the payment
of all expenses incurred by the aforementioned while on their visit.

.... snip ...

.... and then

From: wheeler
Date: 10/21/85 13:41:57
To: (ucb marketing rep)

is berkeley still interested in pitch/discussion on satellites? How
'bout late next week or preferably sometime the week following? There
are some SBS folks coming out the week after next ... and we have
tentative plans on going by Cyclotomics in berkeley ... a company that
has some forward error correcting hardware ... i believe there is a
berkeley professor invovled in the company, E. Berlekamp.

.... snip ..

It wasn't my first trip to Japan ... including being there in the early
70s ... one of the places I got to go as part of my hobby supporting
HONE ... and HONE starting to spread HONE clones around the world.
http://www.garlic.com/~lynn/subtopic.html#hone

in the mid-70s, Endicott con'ed me in to doing 138/148 market planning
trips to various places in EMEA and AFE (including Tokyo).

--
40+yrs virtualization experience (since Jan68), online at home since Mar1970

From: Anne & Lynn Wheeler on 16 Nov 2009 01:42

re:
http://www.garlic.com/~lynn/2009p.html#84 Anyone going to Supercomputers '09 in Portland?
http://www.garlic.com/~lynn/2009p.html#85 Anyone going to Supercomputers '09 in Portland?
http://www.garlic.com/~lynn/2009q.html#0 Anyone going to Supercomputers '09 in Portland?

random other fiber ramblings from long ago and far away; I had been doing
work with NSC & HYPERChannel off & on back to 1980.

Date: 09/21/84 16:59:19
From: wheeler
To: distribution

re: hsdt0918; fyi - we met with various gpd people, lsg, & nsc to
discuss the plans for expanding the use of hyperchannel. We discussed
the various alternatives to T1/A710s at the 4000'-8000' ... would
extended A220<->A220 coax be adviseable. Based on statements about
continuing to expand hyperchannel connectivity in the plant site area
.... the NSC people finally said that a T4/backbone/fiber-optic system
is being beta-tested & is scheduled for next summer. The backbone
system will support a T4/fiber-optic cable to a distance of 50 miles,
off of which can be hung Hyperchannel & Hyperbus interfaces.

.... snip ...

Date: 10/12/84 22:00:50
From: wheeler
To: distribution

re: channel attach interface;

talking to some people in c.s., there is apparently a channel attach
card that was developed in Tucson for testing control units. It is
capable of data streaming at 3 megabytes, fits on a PC card (in a PC),
will cost about $150, and sjr c.s. are suppose to have some by the end
of the month. Drawback is that it is limited to transfer operations of
no more than 8K-byte data blocks.

Now if we can get a 100-280 mbits fiber optic HSLAN on the back end,
we can have host attach packet network ... and also hang DNS off of.

.... snip ...

little later I was also on XTP technical advisory board (although the
communication division non-concurred and claimed that I wasn't
allowed).

Date: Sun, 13 Nov 1988 11:10:30 PST
From: wheeler
To: internal distribution

re: xtp;

What is XTP?

- reliable end-to-end messaging

* traditional stream, bulk data, reliable datagram
* internet routing
* rate-based flow control
* selective retransmission
* message check-sums in addition to physical
layer's crc

- context orientation
* packetization
* reassembly
* out-of-order reception

- datagram support
* arbitrary size
* reliable multicast

- xtp routing
* real-time frame relay
* uses external routing protocol (eg. dod/ip)
* internal xtp routing tables set by
external routing protocol
* xtp adapts to router technology
developments

- xtp reconfiguration occurs when
* gateway redirects
* gatew failure detected

- current applications
* mission-critical applications
* "just-in-time" manufactoring
* a navy survivable adaptable fiber optic
embedded network (safenet) standard
* USAF modular simulator architecture
(boeing aircraft)
* NSA "doo-dads"

- well suited for fiber-optic communication
* NSFnet

- coast-to-coast LAN bridging
* "blazing" SMDS transport

- Metropolitan networks (802.6)

- Establisment backbones
* heterogeneous co-residency
* existing protocols (XNS, TCP/IP, OSI, 802.2)
* Existing hosts/IWSs

.... snip ...

--
40+yrs virtualization experience (since Jan68), online at home since Mar1970

From: Anne & Lynn Wheeler on 16 Nov 2009 10:38

First | Prev | Next | Last
Pages: 1 2 3 4 5
Prev: Performance of SMT on Atom
Next: Lipitor generic cheapest. Buy cheap Lipitor online. Buy Lipitor cod. Lipitor overnight delivery.