rod@LION.ISI.EDU (Rod Van Meter) (09/25/86)
Problem: Suddenly decreased performance from our system disk. We have a cluster of 8 11/750's with an HSC50 controlled RA81 as our system disk. Currently we have a second HSC50 plugged in ready to do dual-porting, but it is not yet enabled. This happened last weekend. At the same time, our apparent access to the system disk was sharply decreased in speed. Using Monitor and information from the "Guide to Performance Management" we determined that our (rough) average current disk access time is from 100 to over 600 milliseconds, depending on the node and load. Does anyone know what kind of access times can normally be expected in a large cluster, and if a nominally inactive but wired in place HSC50 could in anyway affect our access? The transition from fast to slow was sharp, but we have no numbers to substantiate this, just our own impressions and our users' complaints. Interactive work has become difficult, and virtually any request (even DIR or a simple delete) takes 6-12 seconds before it leaves the I/O Request Queue, sometimes even more than a minute. Any ideas? Rod Van Meter USC Information Sciences Institute (213)822-1511
EFanwick.STHQ@XEROX.COM (10/10/86)
First, do you have all 8 750's going against one system disk? I believe DEC recommendation is to have not more the 4 CPU's against a system disk. Next, are your page and swap files on the same disk as your system disk, if so you might think about creating secondary page and swap files on another disk and reducing the size of the primary to the miminum so the system use the secondary files thus spreading out your disk IO. Last, you may have fragmented your system disk, using the command MONITOR IO, observed the WINDOW TURN RATE, if this rate remain above 0 for long periods of time during peak times, this may indicate a disk that is fragmented. If you have SPM then use that to determine the state of your system disks. Eric S. Fanwick Systems Engineer Xerox Corp. EFANWICK.STHQ@XEROX.COM (203) 968-4533