Effect of Partition Placement on Performance

[dhruv17singhal]

Hi,

I have four hard disks and three operating systems (Windows 7, Ubuntu and Fedora) and I want advice on partitioning the disks. In my opinion, I have two options:

1. Install all the operating systems on one hard disk (since only one partition will be in use at a time) and use the other three for different purposes like download storage, media storage, and Virtualbox disk file storage, etc.

Potential Advantage: that the OSes can have concurrent access for different jobs (e.g., Virtualbox, download, and media-serving activities, simultaneously); more space for the data partitions.

Potential Disadvantage: the OSes installed in the latter part of the OS disk will be lower in performance.


2. Install the OSes at the beginning of three of the disks, and the other data partitions on the remaining part of the disks.

Potential Advantage: Better performance of all the OSes, in general.

Potential Disadvantage: Lesser space for data partitions.


How do you think I should use the disk? How much (benchmark results / statistics will be very helpful!) of an impact does the placement (at the beginning or at the end) of a hard disk partition has on performance?

 

[TVeblen]

It had a lot more impact in the past when we were dealing with slower drives, IDE controllers, and 33 or 66Mhz system buses. But with modern systems there is very little real-life performance gains from drive position, even with mechanical spinning hard drives.

I am sure that the speed difference is still there and could be picked up by benchmark software, but the difference in today's machines would be measured in milliseconds, not full seconds.

I used to set up a small partition in the beginning of a second hard drive just to use as a page file in order to boost performance (which seemed to make a noticeable difference at the time. But I did the same thing on my new machine when I first set it up and it made no difference at all. With the speed of today's hardware it is just splitting hairs, IMHO.

 

[dhruv17singhal]

Benchmark Results

I did some benchmarking today and I'm posting my results.

Setup: I created two 100 GB partitions, at the beginning and at the end, respectively, of an empty hard disk.

Software: Crystal Disk Mark

Results:

* MB/s = 1,000,000 byte/s

Note: The first reading is of the partition at the beginning and the second is of the partition at the end.

1. Sequential Read :
   1. 113.778 MB/s
   2. 78.929 MB/s
2. Sequential Write :
   1. 106.878 MB/s
   2. 70.270 MB/s
3. Random Read 512KB :
   1. 47.416 MB/s
   2. 40.233 MB/s
4. Random Write 512KB :
   1. 62.978 MB/s
   2. 43.848 MB/s
5. Random Read 4KB (QD=1) :
   1. 0.651 MB/s [159.0 IOPS]
   2. 0.640 MB/s [156.2 IOPS]
6. Random Write 4KB (QD=1) :
   1. 0.550 MB/s [134.2 IOPS]
   2. 0.481 MB/s [117.5 IOPS]
7. Random Read 4KB (QD=32) :
   1. 1.656 MB/s [404.2 IOPS]
   2. 1.539 MB/s [375.8 IOPS]
8. Random Write 4KB (QD=32) :
   1. 0.545 MB/s [133.1 IOPS]
   2. 0.471 MB/s [115.1 IOPS]

Test : 1000 MB (x5)
OS : Windows 7 Ultimate Edition SP1 [6.1 Build 7601] (x64)

These results surprised me as, like TVeblen, I was not expect such a disparity is data rates.

 

[TVeblen]

Thanks for posting that information, very interesting. The differences are larger than I would have expected. But then you are dealing with a small partition at the very ends of the platter, so I think we can look at the results as the worst case scenario.

 

[dhruv17singhal]

I agree with that. The hard disk was of capacity 1 TB; I think the deterioration in performance would have been far less, if I had used a 320 GB or 160 GB disk.