pete@octopus.UUCP (Pete Holzmann) (05/01/88)
A few people made comments about using lots of small partitions to take
advantage of a big disk drive... This set the jello that's left of my
brain off again. Here's some stuff I forgot to include in my summary:
There are significant tradeoffs involved in deciding whether to parcel
your big disk drive(s) into lots of little (<32MB) partitions or a huge
partition. I'll try to make this simple...
CLUSTER SIZES
Huge partitions *require* big cluster sizes. On average, you waste 1/2
of a cluster of disk space per file ('cuz the last cluster in each file
is, on average, half full). If you have lots of files, you therefore
waste a chunk of disk space by going to big clusters. On the other
hand, DOS does a linear traverse of the cluster list for a file every
time you access the file. Big files get lots of clusters. Therefore,
big clusters can *significantly* speed up access times for big files.
MEMORY
To begin with, DOS3.3 is a hog compared with earlier versions. Usually it
takes up around 40K extra RAM from your 640 (compared with earlier DOS
versions).
Now add N buffers **PER PARTITION** and you've got a lot of space used up
by DOS. If you have partitions C: through J: (8 partitions) and you set
'buffers=10' in your config.sys, you've just allocated 10*8*512 = 40K worth
of buffer space. I've got a customer with 480MB of disk space; he tried
lots of small partitions with buffers=20 (for better speed). That came
out to 16*20*512 = 160K of buffer space! And he wondered why AutoCAD wouldn't
load. [I don't remember offhand whether you also get a set of buffers per
floppy. You can play with this by changing N in your config.sys, then
comparing the change in memory usage.]
ORGANIZATION
This could be considered a wash. Since you can SUBST a drive letter for
a path on another drive, you can make big and small partitions look pretty
much the same to a user.
On the other hand, I personally like to have everything organized under a
single hierarchy; this is tough when things must be spread over lots of
partitions.
Note that if you have BIG files, it is a pain to go looking
for them in lots of different disk partitions. Some folks toss 1MB+
files around with impunity. It doesn't take much to fill up a 32MB partition
with stuff like that!
COMPATIBILITY
2K cluster sizes (and therefore, small partitions) are in general more
compatible with software utilities that play with low level hard disk
stuff. Most end users will never notice the difference, however; they
don't regularly reformat their drives!
CONCLUSION/RECOMMENDATIONS
If you have memory to spare in your 640K, and you don't need room for
big files in a partition, then small partitions will probably be fine.
If you are playing with big databases, you'll benefit a lot if you make
a partition for the database file that has big clusters (e.g. 32K). Do
this even if you make the partition itself small. Going from 2K to 32K
clusters can give an order of magnitude increase in performance!
If neither of the above applies to you, then what do I care? :-) Make
up your own mind; it's your system, not mine!
Pete
--
OOO __| ___ Peter Holzmann, Octopus Enterprises
OOOOOOO___/ _______ USPS: 19611 La Mar Court, Cupertino, CA 95014
OOOOO \___/ UUCP: {hpda,pyramid}!octopus!pete
___| \_____ Phone: 408/996-7746mvolo@ecsvax.UUCP (Michael R. Volow) (05/03/88)
A previous poster on this subject mentioned that even if one had a large partition (e.g., 30 M) with 2k cluster size, additional partitions smaller than 10-15 M would automatically default to 4k cluster size (what a waste!). To see if this was true on our Zenith 248 (AT clone) running Zenith MS DOS 3.20 with a 40 MB hd, I found that a 2-byte file in the 30 MB partition did in fact tie up 2K, but a 2-byte file in the remaining 10 MB partition tied up 4k!!! What a bummer! The poster (I forget who) suggested a modification to correct this, but I am loathe to mess directly with the partition table or the FAT. Is there any simple way to format small partitions for 2k clusters? Michael Volow, M.D. Dept of Psychiatry, Durham VA Medical Center, Durham, N.C. 27705 919 286 0411 mvolo@ecsvax.UUCP