Partitions and Speed

[whitewale]

Usually I use two partitions on most drives over 40 GB. Now I'm switching to a 74 GB Raptor for faster disk access, and realize that I have no idea what multiple partition on a physical drive do to access speed.
I have an older 80 GB IDE drive for documents I plan on reusing, so I don't need th second partition per se.
Is there any physical reason to go with one or two partitions if fast program load times is the goal?

 

[Ice Czar]

that would depend

http://www.hardforum.com/showthread.php?s=&threadid=741512

 

[whitewale]

well, I read through all that even before I posted this, and I'm still not sure.

 

[BillR]

that would depend

http://www.hardforum.com/showthread.php?s=&threadid=741512

A very good read. My “Profound truth” with my 74 was / is I tried a smaller partition for the OS and used the rest for programs. Then, out of shear boredom I just made it one big partition.

I’d have to say that in every day use I saw no performance difference with the applications I use.

I’m not a benchmark maven as I find that benchmarks and real world performance have very few correlations.

I do find that placing things like the cache files from IE on a different drive help a lot, far less fragmentation.

I have tried moving the swap file to a different partition and while that seemed a valid worthwhile move in Win 9X I find less and less benefit to it with XP.

As a rule of thumb I try to keep disk and or partition size at not much bigger then 80 gig or so just for management purposes. Defragging a 250-gig drive can be a major pain

$.02 worth of opinion at your service

 

[Ice Czar]

well, I read through all that even before I posted this, and I'm still not sure.

without a "typical" access pattern I cant say either
it really depends on whats on the partitions
by limiting the size of the partition on the OD (outerdiameter) which as an example has the OS, the possibility that a system file that is needed in a day to day situation would thus be closer to the average positi0on the head is already in, so it has fewer degrees of arc to travel and likely less access time

whereas if it was on one large partition its possible that it could endup further away on average and increase the access time

an the other hand, you would increase the access time to data on the second partition on average

the trick of course is determining "typical" and "average"

cheating counts though so adding drives on seperate channels
and putting highly accessed things on the outer partitions of them
with rarely accessed data on the inner partitions is a viable "general answer"
since by doing that you generally get concurrent access

but on a single drive obviously its more complex
and either boils down to trial and error with your personal subjective judgement
or some serious logging

I say subjective, because while other things you dont notice much (and it could be alot of things)
might speed up, you can end up focusing on one specific thing that now lags
thats why you get such diverse opinions whenever you start talking one strategy over another

 

[whitewale]

So basically what you're telling me, for optimal performance I shouldn't have more than maybe 5 gig per drive (that should stay nicely in the outer sectors), and what I really need is a second 74 raptor? One for the OS only, and one for my Everquest habit?

 

[Octave]

yes.....

 

[Ice Czar]

So basically what you're telling me, for optimal performance I shouldn't have more than maybe 5 gig per drive (that should stay nicely in the outer sectors), and what I really need is a second 74 raptor? One for the OS only, and one for my Everquest habit?

ahh the good old days

The effects of destroking
To reduce access time, drives can be destroked. This process enables data to be stored only on the outer edge of a disk; for more information about destroking, see the "Optimizing systems for performance" section of this article. If the 14 Cheetah 10K drives were destroked so that they had the same capacity as nine Cheetah 15K drives, the Enterprise Systems Lab estimated that nine or more Cheetah 15K drives could provide greater total capacity and better performance than the destroked 10K rpm drives. To prove this hypothesis, the lab tested the performance of 14 Cheetah 10K drives destroked to a total capacity of nine 36 GB drives. The performance results show that, even when destroking the Cheetah 10K, the Cheetah 10K requires 40 percent more drives to deliver 2 percent fewer IOPS, with 3 percent slower response time and 10 percent less capacity, than 10 Cheetah 15K drives (see Figures 6 and 7).[/QUPThe effects of destroking
To reduce access time, drives can be destroked. This process enables data to be stored only on the outer edge of a disk; for more information about destroking, see the "Optimizing systems for performance" section of this article. If the 14 Cheetah 10K drives were destroked so that they had the same capacity as nine Cheetah 15K drives, the Enterprise Systems Lab estimated that nine or more Cheetah 15K drives could provide greater total capacity and better performance than the destroked 10K rpm drives. To prove this hypothesis, the lab tested the performance of 14 Cheetah 10K drives destroked to a total capacity of nine 36 GB drives. The performance results show that, even when destroking the Cheetah 10K, the Cheetah 10K requires 40 percent more drives to deliver 2 percent fewer IOPS, with 3 percent slower response time and 10 percent less capacity, than 10 Cheetah 15K drives (see Figures 6 and 7)

Optimizing systems for performance

To ensure that the disk drive is not a bottleneck in the system, the drive's access time must be minimized. A technique called destroking , or short-stroking , allows administrators to reduce the access time of hard disk drives. Destroking formats the disk so that data is stored only on the outermost part of the disk; therefore, the arm does not need to seek very far to reach the right track.

In Figure 13 , data is stored on the outer diameter (shown in dark blue) of a destroked disk. The lighter blue area represents available data storage that will remain unused. Destroking can compensate for the longer latency and seek time of a 10K rpm drive. However, the result is an expensive cost per usable gigabyte and cost per transaction to the end user.

The Cheetah 15K drive decreases access time without sacrificing disk capacity. Thus, fewer drives of a given capacity are needed to meet the performance requirements of the system if those drives are Cheetah 15K drives instead of the Cheetah 10K drives.

For example, assume an administrator wants to store 360 GB of data and wants to minimize access time to that data. If the administrator stores this data on 10 Cheetah 15K drives, the average seek time plus latency is 5.6 ms. To maintain this 5.6 ms figure on a set of Cheetah 10K drives, the seek time must average 2.6 ms because the latency of the 10K rpm drives is 1 ms higher than that of 15K rpm drives. Consequently, the administrator must destroke the 10K rpm drive to drop its seek time to the 2.6 ms seek-time limit, resulting in reduced capacity for each 10K rpm drive. The administrator would need additional 10K rpm drives to store the 360 GB of data.

what Im saying is, there certainly would be disk performance to be had that way
is disk accesses your bottleneck?

or do you need more RAM or Faster CPU?

what I described is "fake" destroking, in that the inner area is simply another rarely accessed partition or partitions, Id say if you had more drives, take advantage of the concept in yourt partitioning strategy, but unless you actually are lagging do to disk accesses (which with a typical single user accesss and raptor is unlikely) dont go out and buy another youd probably be disappointed as "performance" is often subjective, youd likely see more benefit from more RAM, its accessability is several orders higher than any disk access, assuming what you want is in memory of course

 

[whitewale]

is disk accesses your bottleneck?

or do you need more RAM or Faster CPU?

I'm now using a p4 3.2 C on an i875 board, 2 GB of cas2 Patriot, 74 GB Raptor and a 6800GT. The old system was an i845 board with a 7200 rpm IDE, and swapping MB and harddrive gave me about a 30% loading time advantage. Since going from 1 GB to 2 GB RAM did very little for the system (unlike 512 to 1 GB what was huge) didn't do anything I don't think more ram is doing a lot.
There is an A64 FX with SLI ultras in the future, but that's such a complete system rebuild that it will have to wait a bit.

 

[Ice Czar]

well with 2GB your certainly not constrained
(I actually saw a big jump in performance from 1GB to 2GB,
but then its was all graphics aps w\ large .tiff\.raw oriented too)

again dependent on what your doing
employing "fake destroked" multiples could make a big percieved difference
or not