Effect of Physical Partition Size on Performance
ITEM: RTA000088256
Is there any information
as to how a smaller or larger Physical Partition
size affects file system performance (or Logical Volume performance)?
If I had a huge file system, however,(say around 2GB)
would it not be more efficient to use 256MB PP's?
Are there any benchmark stats on this and a documented reason why 4MB
was chosen as the default?
ANSWER
There is no relationship between the LVM partition size and disk I/O
efficiency with regard to speed. This is because none of the system
components or application programs depend on partition size to
accomplish I/O. More specifically, the only speed factors that affect
I/O are a) physical location on the disk b) size of each transfer c)
physical speed of the device (which is a hardware limitation).
The factors that affect a) are
- How contiguous are the LVs? The administrator controls this when
the LV is created. If one presumes that a fresh disk is being used to
create LVs, then one can create contiguous LVs regardless of partition
size.
If the disk contains LVs that have been randomly extended so that the
partitions are not very contiguous, then smaller partitions will give
one a greater likelihood of extending LVs contiguously. But, the real
solution is to create then of the right contiguous size in the first
place.
- Placement of the data within the LVs tends to be under random
control of the logical filesystem layer. The user has no control over
exact physical file placement within an LV.
The factors that affect b ) are
- VMM does amounts of I/O based on LRU factors which has no relation
to partition size.
- Programs do amounts of I/O purely based on what they feel like
doing. The I/O sizes have no relation to partition size.
Since I've shown that there is no relationship between partition size
an I/O speed, I would not expect to see any difference in throughput
by varying partition size. I am not aware of any experiments that
were conducted to refute this theory.
The ability to pick partition size was an anticipation by an
insightful designer who realized that disk capacity is rapidly
increasing with time Since the field which holds the total number of
partitions in a logical volume is fixed, it is desirable to not put an
artificial limit on the size of the LV. So, as disks get larger, it
is possible to allow LVs to become larger and not exceed the number of
bits in said field. This technique is commonly done in other
architectures. However these other architectures (Macintosh file
system for example) usually automatically change the partition size
according to the total size of the disk (or logical disk). AIX has
given the administrator the option of choosing the size to accommodate
both large disks and small disks.
Let me do further research on why 4MB was chosen as the default.
From what I can investigate myself , the default partition was chosen
to accommodate maximum capacities of todays disk drives. The pertinent
limit are that the range of partition sizes is from 1Mbyte to 256
Mbytes. There is also a limit of 1016 physical partitions per
physical volume.
If one chooses the minimum partition size, then the resulting size of
1016 Mbytes doesn't even cover the largest of the magnetic drives
(about 2.6 G for 5.25 full height form factor). However, 4G (1016 *
4M) is well within todays largest magnetic drive sizes. But 4M
partitions will be insufficient for optical drives of today or for
magnetic technology of tomorrow. Thus, when drives grow beyond 4G,
one will be forced to use partition sizes greater than 4Mbytes. In
conclusion, partition size is motivated only by physical drive size
and has nothing to do with performance factors.
S e a r c h - k e y w o r d s:
PHYSICAL PARTITION SIZE AIX RISC
WWQA: ITEM: RTA000088256 ITEM: RTA000088256
Dated: 08/1995 Category: ITSAI6000GE
This HTML file was generated 99/06/24~12:43:29
Comments or suggestions?
Contact us