[sclug] Newbie, partitioning 120Gb HDD - recommendations?

Tue Jan 18 20:56:15 UTC 2005

On Tue, 2005-01-18 at 15:51 +0000, Alex Butcher wrote:

> *Modern* discs *don't* have the same number of physical sectors on outer
> tracks as inner tracks, 

Yes I know and it's good to look at your numbers.  So if we know the
boundaries of the zone we can partition the disk on those boundaries.
and choose which partitions to use for which file systems.

Do I put the root file system in the fast zone or do I put home on the
fast zone?

I think the general advice is that you want root to be fast. But is that
the best strategy?  Lets say I want to edit a file.  Well I'm so slow at
typing I guess the speed of the file system is academic.  But vim for
example is about a 0.25Meg but the file I want to edit is say 1 Meg.
(I know vim needs some shared libraries too.)   I have to read in and
write out 1 meg making a total of 2 meg of I/O so I get get the most
benefit from having home in the fast zone.

You really need to know which file system you are doing the majority of
your I/O to before you can make any decision on where to put a file
system.

I just ran hdparm on my system here and got 

/dev/sda5:
 Timing cached reads:   3524 MB in  2.00 seconds = 1764.91 MB/sec
 Timing buffered disk reads:  174 MB in  3.02 seconds =  57.59 MB/sec

[root at localhost ~]# hdparm -tT /dev/sda5

/dev/sda5:
 Timing cached reads:   3424 MB in  2.00 seconds = 1715.69 MB/sec
 Timing buffered disk reads:  484 MB in  3.03 seconds = 159.76 MB/sec

[root at localhost ~]# hdparm -tT /dev/sda5

/dev/sda5:
 Timing cached reads:   3452 MB in  2.00 seconds = 1728.86 MB/sec
 Timing buffered disk reads:  604 MB in  3.00 seconds = 201.30 MB/sec

[root at localhost ~]# hdparm -tT /dev/sda5

/dev/sda5:
 Timing cached reads:   3392 MB in  2.00 seconds = 1698.81 MB/sec
 Timing buffered disk reads:  720 MB in  3.02 seconds = 238.37 MB/sec

[root at localhost ~]# hdparm -tT /dev/sda5

/dev/sda5:
 Timing cached reads:   3424 MB in  2.00 seconds = 1714.83 MB/sec
 Timing buffered disk reads:  438 MB in  3.02 seconds = 144.96 MB/sec

I suspect the slow performance of the first run may be due to the fact
that none of the data was buffered.  What you really need to know is the
raw speed.

I've often had complaints from customers along the lines of.  The
transfer rate of these drives is supposed to be N meg per second but the
best I can get is N/x meg per second.   Some just won't accept that you
can't sustain the N meg per second over long periods of time. 

In the example of the of the 3.8 GB Quantum Fireball from
http://www.pcguide.com/ref/hdd/geom/tracksZBR-c.html lets assume 6
platters thats 12 heads every 696K in Zone one you have to seek to a new
track. In zone 14 then you have to seek after only 366K So is the
improved Transfer Rate due in part to the gain made from not having to
seek? In some ways the numbers don't add up.  (well to me at any rate.)

Zone  0 has 232 sectors per track and a Data Transfer Rate of 92.9
Mbits/s
Zone 14 has 122 sectors per track and a Data Transfer Rate of 49.5
Mbits/s

In one revolution Zone 0 pushes 2.148 more data past the head *but* the
data transfer is only 1.877 times faster.  It might be more reasonably
to expect a rate of 106.3 Mbits/s

It would be interesting to know what the real difference in performance
is.  At the end of the day if your not disk bound it's not going to make
a big difference.  IMHO.

Tom.

> hence:
> 
> # fdisk -l /dev/hde
> 
> Disk /dev/hde: 200.0 GB, 200049647616 bytes
> 255 heads, 63 sectors/track, 24321 cylinders
> Units = cylinders of 16065 * 512 = 8225280 bytes
> 
>     Device Boot      Start         End      Blocks   Id  System
> /dev/hde1               1         499     4008186    c  W95 FAT32 (LBA)
> /dev/hde2   *         500         761     2104515   83  Linux
> /dev/hde3             762        1023     2104515   83  Linux
> /dev/hde4            1024       24321   187141185    f  W95 Ext'd (LBA)
> /dev/hde5            1024        3016    16008741    c  W95 FAT32 (LBA)
> /dev/hde6            3017        5009    16008741   83  Linux
> /dev/hde7            5010        8994    32009481   fd  Linux raid
> autodetect
> /dev/hde8            8995        9238     1959898+  82  Linux swap
> /dev/hde9            9239       24321   121154166   fd  Linux raid
> autodetect
> # hdparm -tT /dev/hde1
> 
> /dev/hde1:
>   Timing cached reads:   1940 MB in  2.00 seconds = 969.18 MB/sec
>   Timing buffered disk reads:  150 MB in  3.03 seconds =  49.58 MB/sec
> [root at caffeine ~]# hdparm -tT /dev/hde1
> 
> /dev/hde1:
>   Timing cached reads:   1944 MB in  2.00 seconds = 971.66 MB/sec
>   Timing buffered disk reads:  144 MB in  3.03 seconds =  47.59 MB/sec
> # hdparm -tT /dev/hde9
> 
> /dev/hde9:
>   Timing cached reads:   1956 MB in  2.00 seconds = 977.66 MB/sec
>   Timing buffered disk reads:  102 MB in  3.03 seconds =  33.68 MB/sec
> # hdparm -tT /dev/hde9
> 
> /dev/hde9:
>   Timing cached reads:   1960 MB in  2.00 seconds = 978.68 MB/sec
>   Timing buffered disk reads:  104 MB in  3.09 seconds =  33.66 MB/sec
> 
> That disc is a recent Seagate ST3200822A 200GByte PATA model.
> <http://www.pcguide.com/ref/hdd/geom/tracksZBR-c.html> shows a 9 year old
> Quantum Fireball disc as having 14 zones, ranging from 232 sectors/track
> down to 122 sectors/track with corresponding transfer rates dropping from
> 92.9Mbits/s at the rim zone to 49.5Mbits/s at the hub zone. A newer 5 year
> old IBM disc also has 14 zones, ranging from 792 spt/372Mbits/s to 370
> spt/172.8Mbits/s.
> 
> Note that the cached reads remain constant at ~970MByte/s, but buffered disk
> reads are ~40% slower by the 40% mark of the disc (I suspect the 40% values
> as being coincidental, rather than directly proportional, otherwise one
> would expect 100% slower at the 100% mark!).
> 
> > Tom.
> 
> Best Regards,
> Alex.