[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Fw: How long will this take?

This means that reads and writes should be on 4KiB boundaries, and writes should be multiples of 4KiB, for optimal performance.  As long as those criteria are met, there's no harm and some real benefits of reading and writing larger blocks than the minimum.

One example benefit, among several possible, is that the OS overhead of one single 1MiB write will be much less than 256 individual 4KiB writes.  (number of system calls performed; overhead inside the OS of merging successive writes to optimize the size of actual disk transfers, etc...)

Hope that helps!

On Mon, Jun 8, 2020, at 9:47 PM, Matthew Campbell wrote:
fdisk said the minimum and optimal access size for my hard drive was 4096 bytes.


-------- Original Message --------
On Jun 8, 2020, 7:42 PM, Dan Ritter < dsr@randomstring.org> wrote:

Jude DaShiell wrote:
> Does any optimal formula exist based on hard drive size which minimizes
> time needed for checking and blanking hard drives in connection with the
> block size value?

If the disk firmware offers it, a SMART long read/verify test
should be close to optimal. Consult smartctl and the disk manufacturer
for details.

For conventional spinning hard disks, the optimal write size would be
a complete cylinder at a time. That varies across the radius of the disk,
and may not be made available to the OS.

In lieue of knowing that, writes which are reasonable integer
multiples of the sector size are very good. 1 MB is probably
good for most drives.

For SMR spinning disks,the optimal write size is one complete
write zone. I've heard that this is standardizing at 256MB, but
I would want to confirm with the manufacturer. There are a lot
of interactions with PMR caches.

For SSD, writing wears out the storage mechanism. A write-all
test won't test reliability; flaws will be detected and remapped
without letting the host know.


Reply to: