Date   : Fri, 07 Jun 1996 11:53:26 +0100
From   : angus@...
Subject: DFS buffer sizes?

Mark de Weger writes:
>  How large are the read and write buffers the BBC DFSs (DFS1.2
>  would be all right; I don't guess there's a lot of difference
>  between the different DFSs) set aside for bget/bput file 
>  operations.

256 bytes. Don't use bget/bput. They have *far* too much overhead associated
with them. Use OSGBPB, then you'll only have the overhead once, for whatever
buffer size you want to use.

Data from my benchmarking tests when writing HDFS:

Times are in seconds. Out indicates file opened for output only, Up indicates
opened for update. Fast indicates using second processor fast bput call. IO
indicates IO processor, Tube indicates second 6502 processor. All tests
include disc spin-up time:

               HDFS    HDFS    HDFS    DNFS
BPUT 32K       Root    Subdir  Acorn   1.20
Out                    54.32   54.36   54.33   55.02
Up                     76.04   75.75   76.28   76.03
Fast, Out              28.80   29.23   28.42   29.41
Fast, Update           49.92   49.75   50.46   50.0


               HDFS    HDFS    HDFS    DNFS
BGET 32K       Root    Subdir  Acorn   1.20
Get                    52.70   52.50   52.71   52.8

 
               HDFS    HDFS    HDFS    DNFS
GBPB           Root    Subdir  Acorn   1.20
Read, IO               29.11   29.17   29.12   29.24
Read, Tube             29.14   29.17   29.11   29.24
Write, IO              29.28   29.40   29.28   29.27
Write, Tube            29.27   29.39   29.27   29.2

The columns are:
       HDFS Root       HDFS format disc, operating in root directory
       HDFS Subdir     HDFS format disc, operating in sub-directory
       HDFS Acorn      HDFS operating on DFS format disc
       DNFS 1.20       Acorn DNFS 1.20 operating on DFS format disc

(HDFS format is *very* similar to DFS format.)

a.
--
Angus Duggan, Harlequin Ltd., Barrington Hall, | INET: angus@...            
Barrington, Cambridge CB2 5RG, U.K.            | FAX: +44(0)1223 873873