________________________________ From: Dave Roeser <[email protected]> To: [email protected] Sent: Thu, March 24, 2011 4:04:43 PM Subject: Re: Region size
Rick, One of my developers and friend, Dave Loveluck, has written a version in C and C++. He has written an assembler version of a balanced Red-Green tree (ok it is Red Black ..). He says writing an assembler version of a balanced binary tree sort would not be difficult - although I need to find out where all this "spare" time is coming from...hmmm Anyway, he wishes to know if this is an "in-storage" sort or is it to use external media? Regards, Dave D Dave: When I was a newbie sysprog hmmm many years ago. Our application people came up with their own sort. I guess it worked fine for them but we did notice a bump in cpu resources that it needed. It probably would not have been if it had stayed that way.. one application program. Then it started to pop up all over their system. Cpu resources were really tight and we really had no place to grow. Another CPU was not an answer we were already an MP shop. I got orders to investigate and had a couple of meetings with the not to friendly programmers. AFter getting rebuffed I went over their heads and only then did I get some co-operation. I finally got some real data I could try out and got the OK to bring in SYNCSORT for a trial (this was just before DFSORT came out IIRC). I also for completeness tried another sort product that just faded away (the name I mean). I did a benchmark and hands down SYNCSORT beat the pants off of their own sort. Just as I was about to suggest that syncsort be ordered DFSORT became available. So, as I wanted to compare apples to apples I got DFSORT in on a trial as well. Well I had issues with DFSORT. I think we were one of the first MP shops that tried it with DFSORT. I had to open a sev 1 PMR because of this. Politics being what they were I was pressured to come up with answers. To try and shorted this SYNCSORT was the cheapest(CPU usage) when it came to sort (story revised later). We got the application people to replace syncsort with their sort and it did help. 5 Years later at another shop using similar data (production) and SMF, DFSORT came out on top. I developed some really nasty data to test sort on and every version of sort that came out and DFSORT won by a small but in our case significant amout (10 percent was typical). The shop was pure COBOL and absolutely zero assembler so I did not have to try and put any bubble sort fires out. Ed ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [email protected] with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
