Warning: Long boring post (hard drive configuration)...
For whomever might be interested, here's the story of my 20 GB
drive install in WinNT, what I learned about boot partitions,
how I finally decided to configure my disks to support the operations
I want, and more thoughts on NT page files.
First, NT (SP4) recognized and formatted the 20GB drive with absolutely
no problem. Second, the 7.8 GB boot partition also worked with no
problem. Good headlines, no? <G>
The following is my final disk configuration. Sizes are in MB = 2^20
bytes and GB = 2^30 bytes (i.e., 1 GB = 1025 MB). You get somewhat
different results expressing things in marketing
MB = 10^6 and GB = 10^9.
My system now has three big disks (all Maxtor Ultra IDE), which in marketing
(10^9) GB units are labeled 20GB (7200 rpm) and 17 GB (5400 rpm).
For grins I tossed in a 2.1GB WDC drive I had on the shelf trying to
find something to do with, which happens to have Win 95, Win 3.1,
and DOS on it, just to see if I could boot it.
Following is the layout from the Disk Administrator:
Primary Master C: Active Primary E: Logical (Extended Partition)
19254 MB NTFS 8032 MB NTFS 11492 MB
Primary Slave D: Active Primary K: Logical F: Logical
16473 MB FAT 2047 MB Fat 2047 MB NTFS 12378 MB
Secondary Master Unformatted G: Logical
16473 MB 1028 MB NTFS 15445 MB
Secondary Slave H: Active Primary Obligatory unformatted
2014 MB FAT 2012 MB 2 MB, courtesy of Win95
Here's how these are planned to be used:
C: Primary NT partition (and boot partition), software installation,
configuration, and obligatory minimal page file.
D: Backup NT boot partition. Also boot DOS and/or Win95 or (probably)
Win98 via bios boot drive selection. I haven't installed these yet
(but can boot DOS and Win95 from H, as it happens).
E: Personal and stock data and databases. My life, in other words.
F: Temp files, software source (downloads, etc), CD images, large data
(scan images), CD-burn stageing. In short, low access/temp access data.
G: Development. (My other life.)
K: Page files, all OSs.
Unformatted: Wiggle room. Can take a bootable OS of any flavor. Might
try Linux or BeOS, who knows?
As an aside, I was indeed able to boot both Win95 and, independently,
DOS 6.22 from the H drive by selecting that drive as the boot drive
in the bios. Of course all the NTFS partitions were invisible, but
I was able to see all of the FAT partitions from DOS, including the
K drive, which was odd because it's a logical drive in a humongous
extended partition. Well, enough of that. Here's the more interesting
stuff (to me, anyhow):
The 8032 MB size of C is the result of experimenting. That was the
maximum size I could specify without getting a Disk Manager warning
(yeah, I was surprised by the warning--DM's getting smarter in his
old age <g>).
The 7.8 GB boundary for booting comes from the Int 13h bios call
for disk I/O, which
has buckets that limit sector addressing to 1023 * 256 * 63, which
at 512 bytes per sector is 1023 * 256 * 63 * 512 = 8,447,459,328
bytes, which in turn is 8056 x 2^20 (8056 MB) or
7.867 x 2^30 (7.867 GB).
However, this is the maximum sector address, not allowing for whatever
NT swipes off the front, not to mention the master boot record.
The size I found by experiment is consistent with 1020 * 256 * 63 * 512,
which is 8032 x 2^20 (7.84 x 2^30).
I stewed a bit about possibly putting the faster 7200 drive as the secondary
master (G) rather than primary master (C, E), but finally decided that when
it comes down to the end, I always wish I had the most space on the C
drive, or here, the other "C Drive" partition.
It wasn't a big part of my decision, but I also did a couple of sloppy
ad-hoc NT boot timings which may be of some interest. In these I found
NT loads about 3% faster from the 7200 drive than from an NTFS partition
on a 5400 drive.
This isn't a very good test of 7200 vs 5400 performance
because by far the bulk of the boot time comes from chkdisk, which
checks all the drives, and from network initialization and (for me)
a domain logon. These things don't depend on boot drive speed versus
other disks, so I would expect actually performance difference to be
somewhat higher.
I also discovered that NT loads about 3% faster again from the
FAT partition on a 5400 drive (or about 6% faster than the
NTFS boot from a 5400 drive). This wasn't a big surprise,
though maybe a little larger than I would have expected,
but it brings me to the subject of page files.
Partly owing to speed and partly to flexibility, I decided to
make the FAT K partition only for page files.
I allocated 2000 MB to the page file on K, essentially the
whole K partition. I guarantee that loss due to 32k
clusters is minimal when you put a 2GB page file in a 2GB partition <G>.
On page files. Further research, hardly definitive but good for me,
has convinced me that the aggregate page file should be as big as one
can afford disk space for within the reasonable bounds that returns
diminish after a while.
This is different from what I was saying a few weeks ago, but
I have not been able to verify any downside to very large
page files other than used disk space, though possibly NT may take a
trifle longer internally to handle page allocation. That would be a
bargain if it cuts down page swapping, however. Another downside
would be fragmentation, but that ain't gonna happen with a preallocated
file in a partition by itself.
I had earlier been concerned about allocating too much real memory to
the file cache based on a large virtual address space, but that fear
was unfounded. File cache is controlled by registry entries and by
default is set as a fraction of real memory. You can change it by
updating the registry, or with a little GUI applet from sysinternals.com
called CacheSet. There's a whole slew of free goodies on this site,
BTW, which has been mentioned here before.
Well, I discovered more, but this is it for now.
Hope some of this helps somebody somewhere <g>.
Spots |
