jps@wucs2.UUCP (James Sterbenz) (06/14/88)
I am interested in experiences involving the installation/operation
of official internet protocols (especially TCP and IP) on large
mainframe systems. I'd like to know how the protocols were installed
(division of processes, what code runs priviliged, etc.), how the
system architecture is utilized (is the communications link on a
communications processor or an IOP), and how the implementation
performs. I'm interested in this for the research issues pertaining to
high performance network-host architectures, rather than from the
standpoint of finding a package to run on my system.
Either pointers to existing documentation (papers, tech reports) or
private communications would be appreciated.
In particular, I am looking for non-Unix, large system implementations,
such as on IBM sys/370 (303x, 308x, 3090) machines under MVS or VM,
Burroughs 6000/7000/A MCP, Univac 1100 EXEC, NCR 8500/8500 VRX, CDC
Cyber 700/800 NOS and NOS/VE, Honeywell DPS 8/80/90 GCOS and MULTICS
(yea I know Burroughs and Univac ane now Unisys, but it messes up the
"BUNCH"), and ICL 2900 VME.
I would guess that there are a number of these installations, especially IBM
(due to sheer numbers) and Univac (due to high government usage).
For example, in the IBM sys/370 MVS environment:
- which parts of the protocols run as seperate tasks, and in what
address spaces
- how is inter adress space communication handeled: via SRB's or using
CMS
- what code runs user state, semi-priveleged, or supervisor state (key 0)
- how are buffers allocated and passed (GETMAIN vs. GETCELL or CPOOL)
- what are performance factors: virtual storage (disk or extended),
real storage utilization, SRM parameters (ICS, IPS), etc.
- are links handeled as communications devices or I/O devices
- if links are I/O devices, was IOS code modified and were new
channel programs written
- if links are communications devices, was any existing VTAM
(or God forbid, TCAM or BTAM :-) code used (actually the effort
might be worth the efficiency gained by using BTAM), or was a
'conventional' access method custom written
- was existing NCP/SNA code utilized in a 3725 or 3705 box or was a
custom control program written
- other issues are important, but this gives the flavor of what I'm
looking for
If responses are e-mailed to me, I will summarize results and post.
--
James Sterbenz Computer and Communications Research Center
Washington University in St. Louis 314-726-4203
INTERNET: jps@wucs1.wustl.edu
UUCP: wucs1!jps@uunet.uu.netjps@wucs2.UUCP (James Sterbenz) (06/14/88)
I am interested in experiences involving the installation/operation
of official internet protocols (especially TCP and IP) on large
mainframe systems. I'd like to know how the protocols were installed
(division of processes, what code runs priviliged, etc.), how the
system architecture is utilized (is the communications link on a
communications processor or an IOP), and how the implementation
performs. I'm interested in this for the research issues pertaining to
high performance network-host architectures, rather than from the
standpoint of finding a package to run on my system.
Either pointers to existing documentation (papers, tech reports) or
private communications would be appreciated.
In particular, I am looking for non-Unix, large system implementations,
such as on IBM sys/370 (303x, 308x, 3090) machines under MVS or VM,
Burroughs 6000/7000/A MCP, Univac 1100 EXEC, NCR 8500/8500 VRX, CDC
Cyber 700/800 NOS and NOS/VE, Honeywell DPS 8/80/90 GCOS and MULTICS
(yea I know Burroughs and Univac ane now Unisys, but it messes up the
"BUNCH"), and ICL 2900 VME.
I would guess that there are a number of these installations, especially IBM
(due to sheer numbers) and Univac (due to high government usage).
For example, in the IBM sys/370 MVS environment:
- which parts of the protocols run as seperate tasks, and in what
address spaces
- how is inter adress space communication handeled: via SRB's or using
CMS
- what code runs user state, semi-priveleged, or supervisor state (key 0)
- how are buffers allocated and passed (GETMAIN vs. GETCELL or CPOOL)
- what are performance factors: virtual storage (disk or extended),
real storage utilization, SRM parameters (ICS, IPS), etc.
- are links handeled as communications devices or I/O devices
- if links are I/O devices, was IOS code modified and were new
channel programs written
- if links are communications devices, was any existing VTAM
(or God forbid, TCAM or BTAM :-) code used (actually the effort
might be worth the efficiency gained by using BTAM), or was a
'conventional' access method custom written
- was existing NCP/SNA code utilized in a 3725 or 3705 box or was a
custom control program written
- other issues are important, but this gives the flavor of what I'm
looking for
If responses are e-mailed to me, I will summarize results and post.
--
James Sterbenz Computer and Communications Research Center
Washington University in St. Louis 314-726-4203
INTERNET: jps@wucs1.wustl.edu
UUCP: wucs1!jps@uunet.uu.net