From: CRDGW2::CRDGW2::MRGATE::"SMTP::CRVAX.SRI.COM::RELAY-INFO-VAX" 11-AUG-1989 07:42 To: MRGATE::"ARISIA::EVERHART" Subj: Re: 8 mm tape drive performance Received: From KL.SRI.COM by CRVAX.SRI.COM with TCP; Fri, 11 AUG 89 04:16:12 PDT Received: from relay.ubc.ca (relay.cdnnet.ca) by KL.SRI.COM with TCP; Fri, 11 Aug 89 04:01:58 PDT Received: by relay.ubc.ca (5.59/1.14) id AA22149; Fri, 11 Aug 89 04:01:50 PDT Date: 11 Aug 89 10:59 -0500 From: John Edgecombe To: Message-Id: <74*edgecombe@ccrs.cdn> Subject: Re: 8 mm tape drive performance >Greetings. I am trying to achieve maximum performance on writes to an >Exabyte 8 mm drive on an MTI controller, attached to a uVAX 3600 (2.7 vup?). >... This report was prepared by Bill Park of Canada Centre for Remote Sensing: Correspondence should be directed to me, as he is not on the net. ------------------------------------------------------------------------ 8mm tape drive performance Status as of Feb. 1989 The EXABYTE-8200 8mm cartridge (246KB/S 2.3GB Capacity) is connected to a DIALOG DU142 controller to a UNIBUS on a VAX-750 running VMS 4.7. The EXABYTE is used for backup and for storing data stripped in real time from a high density instrument tape where for a given run the record size and i/o rates are fixed and maximized subject to the CPU and i/o bandwidths. We have measured the performance capabilities and investigated the peculiarities of the EXABYTE:- 1. Because the EXABYTE drive and not the controller has a 256k internal buffer memory it is possible to exceed in burst mode the maximum time averaged transfer rate of 246KB/sec and there is little interference if several drives are on the same controller. We have achieved burst mode i/o rates of 277KB/sec using QIOWs (inefficient) of 7020byte records. 2. When the tape in mounted or rewound there will be a 30sec delay to bring the tape to the R/W heads before data can be written to the tape. 3. Writing a file mark takes 10secs. 4. There is a startup problem writing data in real time which limits the writes/sec and transfer rate but once this settling in period one can achieve close to the 240KB/sec transfer rate. The table below indicates the effective i/o transfer rate that can be guaranteed for real time applications due to the EXABYTE startup problems and VMS overhead which is thought to be the main problem when the writes/sec is greater than about 25. There are some outriders to the curve in the table below, eg we can transfer about 220KB/sec at 50 write/sec but anything slightly different will fail. We have no explanation for this phenomenon. KB/sec Write operations/sec 250 10 240 15 180 20 100 25 70 30 60 35 40 40 It should be remembered that when we read from the real time device at the same rate as we write to the EXABYTE. 80 i/o operations per second is quite an achievment for a VAX-750! The real time device and the Page 2 EXABYTE HAVE to be on seperate UNIBUS for it all to work. 5. Reading 7020byte records from the EXABYTE using QIOWs we have measured 30 read operations/sec (210KB/sec averaged over many minutes) providing you specify the number of bytes to read as 7020. If you specify 7022 then the reads/sec drops to 14 (98KB/sec)! This phenomenon is also exhibited with a MASSBUS TU78 125IPS CCT drive where the reads/sec drops from 56 to 26 but does not occur when the EXABYTE is driven by a TD controller VIKING UTO where we achieve 22r/sec (154KB/s) max. I have been warned about possible protocol problems with the latter controller but have no details on what these are. 6. Appending or backskipping files or records then attempting to overwrite will not work. 7. At 7020byte records we can store 1.95GB on one cartridge. 8. The measured rewind time is 2 minutes on a full tape. 9. Mount /foreign takes about 1 minute. 10. SET MAGTAPE CCT:/SKIP=END takes 2.5hours on a full tape of 7020byte records. 11. We have experienced no problems using the EXABYTE drives for backup purposes. Status as of AUG. 1989 1. Some changes have been made to the controller and we have upgraded to VMS 5.1. The overall performance of the drive seems to have degraded a few percent. 2. Sometimes the read i/o rate degrades by an order of magnitude for reasons that are not understood. This seems to be cured by turning the computer power off and on. 3. BACKUP/list of a single 157000 block file yields about 100KB/sec -- john John Edgecombe Canada Centre for Remote Sensing Centre Canadien de Teledetection Ottawa, Ontario