Slight variance on 1TB Samsung SSD 850 EVO in my Dell Latitude Laptop

[selwynpolit]

I'm curious why the different results for several different runs:

  +--------------------------------------------------------------------------+
  | ReadSpeed: Hyper-accurate mass storage read-performance benchmark. rel 1 |
  |  Benchmarked values are in megabytes read per second at five locations.  |
  +--------------------------------------------------------------------------+

Driv Size  Drive Identity     Location:    0      25%     50%     75%     100
---- ----- ---------------------------- ------- ------- ------- ------- -------
81  1.0TB Samsung SSD 850 EVO 1TB       483.4   403.0   426.7   432.9   457.6

                  Benchmarked: Friday, 2020-12-25 at 22:49
-----------------------------------------------------------------------------

Driv Size  Drive Identity     Location:    0      25%     50%     75%     100
---- ----- ---------------------------- ------- ------- ------- ------- -------
81  1.0TB Samsung SSD 850 EVO 1TB       510.7   451.2   463.3   466.2   482.4

                  Benchmarked: Friday, 2020-12-25 at 22:51

Driv Size  Drive Identity     Location:    0      25%     50%     75%     100
---- ----- ---------------------------- ------- ------- ------- ------- -------
81  1.0TB Samsung SSD 850 EVO 1TB       516.4   461.8   475.0   476.0   491.6

                  Benchmarked: Friday, 2020-12-25 at 22:52

I was kinda imagining these would always produce consistent results. Do I need to reboot between tests? Perhaps some caching may be going on.

Selwyn

 

[Steve]

Selwyn (@selwynpolit):

What we're learning is that SSDs have a LOT of “stuff” going on beneath the covers. I'll venture that if you were to turn off the machine and run the test again, you would see the results more like the 3rd time. With hard drives — which generally have much less “magic” going on — we tend to see rock solid numbers. And also with many SSD's. But your nice Samsung SSD might actually be changing some of its internals — such as learning which voltage thresholds to use for reading data — as a result of being asked to read.

There's a lot that we still don't know at this early stage.

 

[PHolder]

In the bad old mainframe days, IBM used to rely on a system feature known as DFHSM (aka https://en.wikipedia.org/wiki/Hierarchical_storage_management ). Perhaps the drive, seeing some number of requests for certain sectors, assumes those sectors have become newly important to the end user, and decide to promote them to a position in a hierarchy that makes for faster access. As @Steve said, there really isn't any way for us to know what internal magic Samsung might be using.