Hello Everyone, Thank you all for the answers. It has been something that I just could not get settled in my mind.
Ed Martin Aultman Health Foundation 330-363-5050 ext 35050 -----Original Message----- From: The IBM z/VM Operating System [mailto:[email protected]] On Behalf Of Rich Smrcina Sent: Wednesday, February 17, 2010 11:27 AM To: [email protected] Subject: Re: MDC versus Disk CACHE was RE: eliminating locks in VSE Kris and Marty have great points. Also each workload is different, you may or may not get sufficient benefit from it, depending upon your access pattern and hit ratio (which Kris mentioned). Regarding RECORDMDC, you may want to check the old VSE performance papers by Wolfgang Kraemer, I seem to recall there may have been some resolution to this issue at some point in the past. On 02/17/2010 09:56 AM, Kris Buelens wrote: > MDC works at storage speed; controller cache needs to get out the z. > > Ages ago (but still a bit valid to see gross orders) > Disk = 30 msec > CU Cache= 3 msec > Storage = xx nsec > > An MDC cache hit is also less costly in CPU than a real IO. But, MDC > with a very poor hitratio makes you consume too much CPU. > Extra advantage MDC: when the wanted data are not in the MDC, CP reads > the whole track. Good for sequential processing. bad for random > access. Then you might want to look at using RECORDMDC instead (but, > I thinks that's only for CMS formatted minidisks, no help for VSE). > -- Rich Smrcina Phone: 414-491-6001 http://www.linkedin.com/in/richsmrcina Catch the WAVV! http://www.wavv.org WAVV 2010 - Apr 9-13, 2010 Covington, KY
