John A Pershing Jr <persh...@alum.mit.edu> writes: > Yeah, and it was a poor implementation, Talk about pathlengths! I > worked with a couple guys in Endicott in the early '90s on a > hyper-optimized implementation of SNA on VM. We left VTAM up there in > its virtual machine, and let it do its PU5 stuff like session initiation > and termination. However, we built the routing function into CP: once a > session was set up, data would flow from the wire into CP and be routed > straight up to the target guest, do not pass VTAM, do not collect 100K > instructions. Outbound was even faster, since we transmitted the data > straight from the guest's memory, avoiding a copy into CP's memory. > It's been a long time, but I seem to remember that our send/receive > pathlength was around about 1000 instructions, which matched the > performance of the "native" IUCV-based networking that was being > introduced at the time. Alas, this effort was cancelled. Sigh...
re: http://www.garlic.com/~lynn/2009l.html#13 SNA: conflicting opinions http://www.garlic.com/~lynn/2009l.html#15 SNA: conflicting opinions http://www.garlic.com/~lynn/2009l.html#17 SNA: conflicting opinions http://www.garlic.com/~lynn/2009l.html#43 SNA: conflicting opinions part of the issue was that after future system was canceled ... misc. past posts http://www.garlic.com/~lynn/submain.html#futuresys there was mad rush to get products back into 370 product pipeline (since future system was going to be completely different & replace 360/370 ... there was no point in putting additional effort into 370). part of that involved the mvs group convincing corporate that the vm370 product be killed, the vm370 development group shutdown and all the people moved to POK to support mvs/xa ... or otherwise mvs/xa wouldn't be able to meet its schedule. eventually endicott managed to convince corporate to acquire the vm370 product mission ... but they had to reconstitute a development group from scratch. IUCV was involved in all that. There had been a superset of IUCV deployed on internal systems for quite some time ... but the new group in Endicott decided to re-invent their own. Part of this required IUCV & related mechanisms to go thru a series of enhancments and product releases before they came close to matching the functionality of the original internal implementation. One of my hobbies was doing my own internal product release & support (including HONE) for internal datacenters. some old email discussing move from cp67 to vm370 of lots of my enhancements (mostly during the future system period) http://www.garlic.com/~lynn/2006v.html#email731212 http://www.garlic.com/~lynn/2006w.html#email750102 http://www.garlic.com/~lynn/2006w.html#email750430 including picking up and merging (IUCV superset) special message (from vm370 datacenter in POK). The original special message had been implemented on cp67 at Pisa science center and then later ported to vm370. i had to do something similar in the early 80s when I was doing HSDT project http://www.garlic.com/~lynn/subnetwork.html#hsdt RSCS was basically limited to 56kbit since couldn't get 37xx controller with more than that. RSCS also had some amount of serialization in the vm370 kernel interface ... that could limit RSCS aggregate thruput to 5-6 4k records/second. In HSDT, I had multiple (full-duplex) T1 and higher speed links ... and using RSCS for some of the links, could run into thruput bottlenecks because of the serialization mechanism (maybe 20kbytes-30kbytes/sec). I needed 300kbytes/sec per full-duplex T1 links (and other links required even higher thruput). So in the early 80s, I needed several mbytes/sec. This is recent post going into some detail of replacing that whole serialization mechanism http://www.garlic.com/~lynn/2009h.html#63 Operating Systems for Virtual Machines part of the effort was lifting a whole part of the vm370 kernel implemented in assembler ... moving it into a virtual address space ... re-implementing it in vs/pascal ... allowing the existing synchronous API to work ... as well as allowing asynchronous mode of operation ... and making it run 10-100 times faster (including eliminating all buffer copies). recent post mentioning 37xx 56kbits, fat pipes, and other issues with the communication group (in the early & mid-80s) http://www.garlic.com/~lynn/2009l.html#24 August 7, 1944: today is the 65th Anniversary of the Birth of the Computer from above ... In that time-frame, hsdt project was getting some equipment built to spec ... one friday before I was to leave for trip to the other side of the pacific ... somebody in the communication group distributed an announcement for new online computer discussion group on the subject of "high-speed" communication ... that included the following definition: low-speed <9.6kbits medium-speed 19.2kbits high-speed 56kbits very high-speed 1.5mbits the following Monday morning, on the wall of a conference room on the other side of the pacific was: low-speed <20mbits medium-speed 100mbits high-speed 200-300mbits very high-speed >600mbits ... snip ... -- 40+yrs virtualization experience (since Jan68), online at home since Mar1970 ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html