Lynn,

I didn't get to work with the 3880-13, but with the 3880-23 I think I recall
sequential pre-fetch was initially fetching three tracks, using a
wrap-around buffer to keep track of the last block read and maintaining two
tracks beyond the last track accessed in cache.

With 3990-3 I think this increased to five tracks, and I have no idea about
3990-6 and beyond.

With HDS Arrays the sequential pre-fetch buffering is based on the stripe
length of the underlying parity group and the data length of the start SSCH
that triggers the sequential pre-fetch staging and monitoring. 

Ron

> -----Original Message-----
> From: IBM Mainframe Discussion List [mailto:[email protected]] On
> Behalf Of Anne & Lynn Wheeler
> Sent: Wednesday, March 21, 2012 9:05 AM
> To: [email protected]
> Subject: Re: [IBM-MAIN] megabytes per second
> 
> [email protected] (Ron Hawkins) writes:
> > I'm  afraid sequential pre-fetch kinda of makes your point invalid for
> > sequential IO.
> 
> when ibm first came out with full-track cache (3880-13/sheriff) ... it
> advertised a 90% hit rate ... based on 3380 track, 10 records per track
and
> sequential read, where first sequently read on track would fetch the full
> track ... and then the next 9 sequential reads would already be in cache.
> 
> however, if the application when to full-track buffering ... the same
exact
> application and data would go from 90% hit ratio to zero percent hit ratio
> (effectively each track would be read as a whole, streamed through the
> 3880-13 cache right into processor memory ... and then would have any
> additional cache reference for the track). double "full-track"
> buffering would then overlap retrieval of the following track with the
> processing of the records in the previous track (masking disk retrieval
> latency, aka akin to instruction execution with prefetch &/or out-of-order
> execution to mask processor cache miss latency).
> 
> --
> virtualization experience starting Jan1968, online at home since Mar1970
> 
> ----------------------------------------------------------------------
> For IBM-MAIN subscribe / signoff / archive access instructions, send email
to
> [email protected] with the message: INFO IBM-MAIN

----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO IBM-MAIN

Reply via email to