l...@garlic.com (Anne & Lynn Wheeler) writes: > disk controllers. Moving the 3270 controllers directly off the mainframe > channels ... replacing them with HYPERChannel boxes ... which were much > faster and had much lower channel busy for identical 3270 channel > operations ... resulted in increase in disk i/o thruput.
re: http://www.garlic.com/~lynn/2009l.html#59 ISPF Counter getting the 3270 controllers off the local channels and replacing them with faster HYPERchannel boxes (doing the same 3270 operations) ... reduced channel busy/contention for disk operations and overall system thruput improved 10-15% (w/o noticable degradation for the IMS group system response at the remote site). that much system thruput improvement would easily justify replacing all 3270 controllers with HYPERChannel boxes ... even for local 3270s still in the bldg. screen shot of the 3270 login screen for the IMS group (moved to remote bldg.) http://www.garlic.com/~lynn/vmhyper.jpg old post with earlier analysis of the 3272/3277 versis 3274/3278 terminal response (separate from later emulated terminal measures where 3277/ANR protocol had three times the upload/download rate of DCA/3278) http://www.garlic.com/~lynn/2001m.html#19 3270 protocol part of the issue was that vm/cms with .1sec system response and .1sec 3272/3277 hardware response still met objective of less than .25 second response to the end user. typical TSO with a 1second response (or greater) hardly noticed the significant slowdown moving to 3274/3278 (ykt was touting a vm/cms system with .2 seconds system response ... I had sjr vm/cms systems with similar hardware & workload that had .11 seconds system response). sometime after the STL IMS group was moved off-site ... the IMS FE RETAIN group in Boulder faced a similar prospect (of being forced to use remote 3270s). The move was to a bldg. that was line-of-site to the datacenter ... so T1 infrared/optical modems were used on the roofs of the two bldgs (instead of microwave used for STL) and similar HYPERChannel configuration. There was some concern that users would see outages with the optical modems during heavy fog & storms. For these kind of T1 links ... we put multiplexors on the trunk and defined side-channel 56kbit circuit with (at the time Fireberd) bit error testers (rest of T1 was for 3270 terminal activity). The worst case (in boulder) was a white-out snow storm where nobody was able to get into the office ... which should up as a few bit errors per second. NSC tried to get the corporation to release my HYPERChannel drivers ... but we couldn't get a corporate to authorize it ... so they had to re-implement the same design from scratch. One of the things I had done was to simulate unrecoverable transmission error as channel check (CC) ... which would get retried thru channel check error recovery. Later after 3090s had been in the field for a year ... I was tracked down by the 3090 product administrator. It turned out that customers 3090 (both VM & MVS) were showing unexpected 3090 channel errors (something like aggregate total of 15-20 channel errors across the whole 3090 customer base for the year ... instead of only 4-5). The additional errors turned out to be because of HYPERChannel drivers (on various customer VM & MVS systems) simulating channel check. After a little research ... I determined that the erep path for IFCC (interface control check) was effectively the same as CC ... and convinced NSC to modify the drivers to reflect simulated IFCC instead of CC. -- 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