re: http://www.garlic.com/~lynn/2018c.html#77 z/VM Live Guest Relocation http://www.garlic.com/~lynn/2018c.html#78 z/VM Live Guest Relocation http://www.garlic.com/~lynn/2018c.html#79 z/VM Live Guest Relocation
note that US consolidated HONE (branch office sales&marketing support) systems running SSI in Palo Alto had eight 2-processor POK machines ... "AP", only one with channels ... so had channel connectivity for eight systems with twice the processing power. HONE apps were heavily APL applications so they needed max. processing power ... with relatively heavy I/O. Problem putting larger numbers in the complex was disk connectivity, IBM offered each disk connected to string-switch which connected to two 4-channel 3830 controllers (maximum of eight systems). Part of my wife's problem with POK growing resistance to increasingly sophisticated loosely-coupled (cluster) was burgeoning cluster vm/4341s (both inside ibm and at customers). vm/4341 cluster had more aggregate processing power than 3033, more aggregate I/O and more aggregate memory, for less money, lower environmentals and much smaller floor space. In Jan. 1979 I was con'ed into doing LLNL benchmark on engineering 4341 (before customer ship) that was looking at getting 70 4341s for compute farm (leading edge of coming cluster supercomputing tsunmai). Inside IBM, there was big upsurge in budget for internal computing power ... however dataceenter floor space was becoming critical resource ... vm/4341 clusters were very attractive alternative to POK 3033. vm/4341 also didn't require raised floor along with FBA 3370 and could be placed out into departmental areas ... customers (and IBM business units) were acquiring 4341s hundreds at a time (leading edge of distributed computing tsunami). The cluster 4341s and departmental 4341s were addressing the raised floor bottleneck (both at customers and inside IBM). email from long ago and far away with extract from "Adessa" newsletter Date: 08/26/82 09:35:43 From: wheeler re: i/o capacity on 4341; from The Adessa Advantage, Volume 1, Number 1, October 1981., Strategies for Coping with Technology: ... as of this writing, for roughly $500,000 you can purchase a procssor with the capacity to execute about 1.6 million instructions per second. This system, the 4341 model group 2, comes with eight megabytes of storage and six channels. Also at this time, a large processor like the IBM 3033 costs about $2,600,000 when configured with sixteen megabytes of memory and twelve channels. The processor will execute about 4.4 million instructions per second. ... What would happen happen if the 3033 capacity for computing was replaced by some number of 4341 model group 2 processors? How many of these newer processors would be needed, and what benefits might result by following such a course of action? ... three of the 4341 systems will do quite nicely. In fact, they can provide about 10 per cent more instruction execution capacity than the 3033 offers. If a full complement of storage is installed on each of the three 4341 (8 megs. at this time) processors then the total 24 megabytes will provide 50 percent more memory than the 3033 makes available. With respect to the I/O capabilities, three 4341 systems together offer 50 per cent more channels than does the 3033. .. The final arbiter in many acquisition proposals is the price. Three 4341 group 2 systems have a total cost of about $1.5 million. If another $500,000 is included for additional equipment to support the sharing of the disk, tape and other devices amoung the three processors, the total comes to $2 million. The potential saving over the cost of installing a 3033 exceeds $500,000. - - - - - - - - - - - - - - - - - - - - - - - - - of course Adessa offers a VM/SP enhancement known as Single System Image (SSI) ... making it possible to operate multiple VM machines as a single system. ... snip ... note Adessa company specialized in VM/370 software enhancements, and included some number of former IBM employees. However, live migration implementation was still limited to a few (virtual-machine based) commercial online service providers (original two were spinoffs of the ibm cambridge science center in the 60s). trivia: IBM San Jose Research had also done vm/4341 clusters implementation ... but lost to VTAM/SNA (battle my wife got tired of fighting) ... cluster operations that had been taking much less than second elapsed time become over 30 seconds with move to VTAM/SNA (my wife also had enhancements for trotter/3088, eight system CTCA that reduced latency and increased throughput, but couldn't get it approved). Note that 3033 was quick&dirty effort kicked off after the failure of FS (along with 3081 in parallel) ... initially 168-3 logic remapped to 20% faster chips ... various tweaks eventually get it to 4.4-4.5MIPS. 303x external channel "director" was 370/158 engine with the integrated channel microcode and w/o the 370 microcode. The engineering 4341 in the (San Jose) bldg. 15 product test lab ... with a couple tweaks was used for 3380 3mbyte/sec datastreaming testing ... something that wasn't even remotely possible with 3033 (& 303x channel director). There is Endicott folklore that POK was so threatened by 4341s that at one point they convinced corporate to cut in half the allocation of critical 4341 manufacturing component. -- virtualization experience starting Jan1968, online at home since Mar1970 ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [email protected] with the message: INFO IBM-MAIN
