Radoslaw, Actually this was my experience before I ever joined HDS or went near a lab. Of course we all have different experiences - what a boring world if we did not.
The practice is that sequential pre-fetch from FCAL Disk in RAID-5 or RAID-10 is much faster than the write rate to any tape technology on the market right now. Reads come from disk cache and writes go to a tape buffer. You should be seeing your sequential read cache hit rates operating in the 95-100% range. If you aren't getting this on RAID-5 or RAID-10 then you probably have a disk contention problem. The theory for HDS is that sequential pre-fetch reads from 8 spindles concurrently at the first sequential request loading the whole stripe in to cache - that's about 3.7 CYLS for RAID-5 and 4.2 CYLS for RAID-10. Pre-fetch will then read ahead for another 3.7/4.2 CYLS concurrently from 8 spindles while your disk-2-tape jobs writes to just one tape drive. A single 10K RPM disk can easily handle a sequential, large-block read rate of 40-50MB/sec where there's a seek for every request - faster in lab conditions. So RAID-5 or RAID-10 can feed the cache significantly faster then FICON can transport - best case about 160MB/sec. I don't know of a FICON tape drive available today that can sustain 160MB/sec writes - but I do know a few disk products that can (not just HDS). BTW my experience with ExHPDM has been with ESCON based Symmetrix 7700E, RVA, and RAMAC writing to Timberline drives. For RVA and RAMAC there was a clear and obvious advantage, whereas for 7700E and Symmetrix there was little or no advantage. BTW, check out the restore time from a ExHPDM tape - that can be an eye opener! > > Ron, > > That's contrary to my experience. Maybe in laboratory conditions, the > fastest DASD box, without any other workload could work faster, but I > never observed it in real world. > Jaguars are really fast. Typically for tapes, it can be observed on long > runs. "Sprint" is still DASD domain. > > -- > Radoslaw Skorupka > Lodz, Poland > ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
