Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!ames!decwrl!shlump.nac.dec.com!shodha.dec.com!alan From: alan@shodha.dec.com ( Alan's Home for Wayward Notes File.) Newsgroups: comp.unix.ultrix Subject: Re: 3rd Party SCI Disks Summary: rough KDB50 performance. Message-ID: <389@shodha.dec.com> Date: 23 Aug 89 16:36:12 GMT References: <15166@dartvax.Dartmouth.EDU> Organization: Digital Equipment Corp. - Colorado Springs, CO. Lines: 42 In article <15166@dartvax.Dartmouth.EDU>, steve@avalon.dartmouth.edu (Steve Campbell) writes: > We are considering purchasing a larger DEC RISC processor (5810), and > are thinking of buying 3rd party disks. Emulex offers a disk stack > containing their SD891 drives, which are CDC drives with an Emulex > module attached. Emulex claims that the result is an MSCP/SDI- > compatible unit that can be controlled by the KDB50 just like a DEC > drive such as an RA90. Would anyone care to comment on whether this > will work under Ultrix 3.X? If not, why not? It may work as a data disk. It may not work as a system disk. Why? Ultrix only supports DEC drives. If 3rd party drives work, fine; if not, take it up with the 3rd party. Previous experience with V3.0 and the VAX has shown that VMB doesn't get along with 3rd party emulations of MSCP disks. The boot path on the 58xx may not be as limited, but there is some advantage of putting the system disk on a supported disk. > > We have also heard that although the KDB50 can nominally run up to 4 > disk drives, there is a performance penalty if more than 2 are connected. > Again, can anyone comment? Any specific figures on the penalty? > Doing a simple sequential read of a large file on an RA90 connected to a KDB50 on my VAX 8800 I can get around 800 KB per second (one disk). When I add a 2nd disk the per disk goes to around 450, and the total to around 900. For three disks the total averages in the upper 900's. With all four disks going it's about 1000 KB/sec. In each case the "per drive" bandwidth was in proportion to the number of drives. The performance you see will depend on the application. If there isn't much seeking going on you're doing sequential reads of large files then the numbers I gave may be useful. There were the results of an exceptionally simple minded experiment that I ran while composing this follow. They should be taken as an official performance study given by DIGITAL. Your mileage may vary depending on the configuation and application. -- Alan Rollow alan%nabeth.dec.com@decwrl.dec.com