nagy%warner.hepnet@LBL.ARPA.UUCP (06/03/87)
>Dave Neiman Csnet: DNeiman@carleton.edu >Carleton College uucp: ...{decvax|ihnp4}!stolaf!ccnfld!dneiman >Northfield, MN 55057-4040 (When in operation) >We are considering the purchase of a 6250 bpi tape drive, which will replace >either or both of our current tape drives. My questions: >Is is possible from the above information to determine whether and how much >the 6250 tape drive will speed up backup (Is the bottle neck tape throughput, >disk i/o, or cpu?). With 1600 bpi backups of 5 RA81s, I'd guess your major bottleneck is tape hanging (i.e. operator) time. Moving to 6250 tapes, you will probably find that either CPU time or disk accesses (head motion) is the limiting factor. >More info: This drive may or may not be placed on the HSC; it may go on >one of the other machines. Is a 750 a reasonable place for a 6250 bpi drive? >How about a 780? What are the performance tradeoffs? According to the Digital tape people, the 750 does not have enough CPU power to drive any 6250 bpi drive at anything near full capacity. This means that TU81's/TA81's will not stream on 750s. The 780 has just enough CPU power to stream the Tx81, though I remember the Digital people say its right on the edge. THe problem can be alleviated by writing your 6250 bpi tapes /NOCRC (NOTE: DO NOT DO THIS ON 1600 bpi tapes as the hardware is not writing its own CRCs at 1600 bpi). So, either connect the tape to the 780 or the HSC and do your backups from the 780. >Currently, backup is done on both tape drives simultaneously, performing >incremental backups on different disks. Backup is usually done in the evening, >when the administrative machines are less loaded. We may purchase a sixth >RA-80, which will be backed up identically as the others. Will the use of a >single 6250 bpi tape drive make backup slower overall? How much will it >compensate for the simultaneous use (on different cpu's) of two 1600 bpi >drives? The VAXCluster I used to manage has 3 785's and an 8650. The disk farm for the development side is 7 RA81s. The daily incremental backup for the 7 RA81s (and 1 RM80) is done as 6 separate volumes as 6 separate save sets onto a single 6250 bpi tape. This usually takes 10-20 minutes during normal operations (8-10 AM); having it all fit onto a single tape is nice. For full backups, the 6250 bpi drive will reduce the number of tapes needed by a factor of 4; thus the number of tape hangs, etc. This alone can have a sizeable effect in reducing your backup times. On the other side, since you are likely to be disk-head-motion limited, doing only one disk at a time will slow things down, but the data can now go onto the tape at a 4 times greater rate (i.e., you are probably not likely to be limited by the tape speed unless you get a streamer and cannot keep it streaming) which will compensate somewhat. When we made the switch from 1600 bpi to 6250 bpi, our backup time went down by slight more than a factor of 4 for doing full backups. Basically it went from 10 hours or so down to about 2 hours and made our standard schedule of weekly full backups reasonable once again = Frank J. Nagy = Fermilab Research Division EED/Controls = HEPNET: WARNER::NAGY (43198::NAGY) or FNAL::NAGY (43009::NAGY) = BitNet: NAGY@FNAL = USnail: Fermilab POB 500 MS/220 Batavia, IL 60510
bobp@tekcae.TEK.COM (Robert N. Perry (Bob)) (06/04/87)
-- Robert N. Perry (Bob) Tektronix-Beaverton, Oregon "Skydive for better health" bobp%tekcae@tektronix.TEK.COM "When I works, I works hard. When I sits, I sits easy. When I thinks, I goes to sleep."