From: lederman@star.enet.dec.DISABLE-JUNK-EMAIL.com Sent: Wednesday, July 26, 2000 7:03 AM To: Info-VAX@Mvb.Saic.Com Subject: Re: GS140 write I/O wall I haven't seen any 'walls' on I/O performance. I passed this message around our group, and got the following comment from one of my co-workers who has done a lot of I/O testing. There must be something else going on in the original configuration that wasn't mentioned in the original posting. More information on exactly how the system is configured, what application is being used, how I/O is being measured, etc., would be needed to tell what is really going on. ======== I have been running on an 8 CPU GS140 for several years now. The system has evolved from EV6/525 and EV6/700. My performance tests have delivered maximum WRITE throughputs varying from 1.2 Megabytes per second (1 block random write IOs ) to 94.3 Megabytes per second (127 block random write IOs )to 150 Megabytes per second (127 block sequential write IOs ) 248 Megabytes per second (127 block sequential READ IOs ) with various configurations including CI, SCSI and FC. I may have come close to being CPU bound, but analysis of data indicates that adapter and controller capacity were the limiting factors. My performance was achieved through artificially configuring the tests to distribute the load as evenly as possible across all system resources involved. Thus you cannot take these numbers for granted since performance is dependent upon all of the configurations and settings on your system in the controllers, adapters, and operating system AND applications and users. I am not saying that anyone can achieve these levels of performance with any CI, SCSI or FC configuration, but I have not encountered a bottleneck caused by the platform itself. - Mark DeYoung Mark.DeYoung@Digital.Com OpenVMS Performance Expertise Center -- B. Z. Lederman Personal Opinions Only Posting to a News group does NOT give anyone permission to send me advertising by E-mail or put me on a mailing list of any kind. Please remove the "DISABLE-JUNK-EMAIL" if you have a legitimate reason to E-mail a response to this post.