On 19/10/2016 22:50, The Natural Philosopher wrote:
On 19/10/16 22:40, John Rumm wrote:

Partition misalignment is a problem on *any* AF drive.

I give up.

So in summary, in the blue corner we have:

Loads of technically clued up people dealing with partition alignment
problems, and lots of them publishing results of the performance gain
when they are fixed.

We have the makers of software tooling, providing specific capabilities
in their software for dealing with the problem. We have drive makers
promoting and linking to said tooling.

We have technical notes from the drive makers explaining the problem.

We have good old common sense that should be bleeding obvious to any
developer, that when you are living in a 4k sector world, you optimise
your code for that, even if it means that 10% of the time you will get a
negative impact from non 4k aligned accesses.

We have OS vendors specifically making changes to their partitioning
tools to support 4k alignment (including the holy tribe of Linus).

We have tech notes like:

https://blogs.technet.microsoft.com/...isks-and-ssds/

Including things like:

"Windows 7 will add additional benefit to the use of SSDs, as it
configures the partition with an disk alignment, that is optimized for
best performance. "

_and_

"€“ Set partition alignment:
NAND flash storage devices are functionally different from traditional
hard disk drives in three ways: physical structure, logical structure,
and bit erase operation. The logical structure of SSDs differs from
tradition hard disk drives because they are built upon pages, which are
a given number of bytes (typically 2 KB or 4 KB), and blocks, which are
a given number of pages. When the user wants to erase (re-program or
write-erase) a given set of bits in any block, the entire block must be
erased. This particular anomaly is directly associated with the physics
of how NAND works at the die level. If an operating system or data
partition is not properly page aligned, there are undesirable
consequences. Specifically, if the offset value where the partition
begins does not line up with a given page boundary, the SSD flash
management system is burdened with extra overhead each time there is a
request to perform an erase data operation.

If you wish to use the align parameter in the DiskPart tool, the correct
value should be 1 MB for SSDs with page sizes up to 4 KB, and 2 MB for
SSDs with a page sizes from 4 KB to 16 KB."


_and_

https://blogs.msdn.microsoft.com/jim...l-server-2014/

"Partition alignment remains a best practice for all versions of Windows
Server as well as SQL Server, including SQL Server 2012 & SQL Server
2014. No exceptions. Period. If, for whatever reason, misaligned
volumes are created, they will fail to deliver their expected
performance. SQL Server installed on such volumes will suffer
concomitant performance degradation."

and just in case you thought there was a free pass for SSDs:

"The performance impact of misalignment is not as apparent on SSD
relative to spinning media. *Yet partition alignment is required for
optimal performance.*" [my emphasis]

Come back when you actually understand how an SSD works.

Then in the red corner we have someone who; with no supporting evidence,
no benchmark results, and no experience of actually writing SSD drive
firmware, claims this is all a figment of the collective imagination,
any only his way is the one true and righteous one. We just have to have
faith!


Sorry, but positions for quasi religious storage guru are now closed.


--
Cheers,

John.

/================================================== ===============\
| Internode Ltd - http://www.internode.co.uk |
|-----------------------------------------------------------------|
| John Rumm - john(at)internode(dot)co(dot)uk |
\================================================= ================/