Value of SSD vs SD card for GoPiGo3 robots

GoPiGo
1

This article seems to suggest that a mostly read-only robot might not benefit (or perhaps even be penalized) having a SSD on-board.

I tested Dave’s current SD card (using the CLI version of the PiOS desktop diagnostics):

NOTE:  Sequential write speed can vary by 7% from run to run,
       but Random r/w speeds will vary by less than 1%

RESULTS:

*** 32GB ROS2HH SD card (in micro SD extender)
Sequential write speed 21607 KB/sec (target 10000) - PASS
Random write speed 645 IOPS (target 500) - PASS
Random read speed 2497 IOPS (target 1500) - PASS

In my circumstance: Using the SSD was

  • 28% faster than SD card for random read, and
  • 10x faster for random write:

(The SSD test was on a Pi4 4GB processor that was not mounted to, nor powered by a GoPiGo3, with SSD powered through the same USB-A USB3.0 data connector)

2 Likes
2

There’s no such thing as a “read mostly” operating system, even with “atime” disabled.

You can, and should, consider any operating system a high write-pressure environment.

However, I have found by my own subjective tests that using a SSD on my 'bot made enough of a difference that it was noticable.

2 Likes
3

Also invalidates all my battery life studies and my power distribution design if this is true:

mSATA SSDs: These devices have a power consumption range of 0.21-1.20 watts when idle, 
2-5 watts when reading data, and 5-8 watts when writing. 

M2 SATA SSDs: These SSDs consume between 0.30-2 watts in idle mode, 
2-6 watts when reading, and 3-9 watts when writing data
2 Likes
4

SSD’s do consume more power.

USB SSD devices can consume considerable amounts of power.

However, the key word here is “can” not “will” or “must”.  It really depends on your use-case.

Should you re-profile your battery studies?  Absolutely!

Should you get all worked up over it?  I’d say “no!”, unless and until you have a sufficiently bad situation, and even then I would strongly recommend against it.

1 Like
5

Initial tests show no consistent difference in SD card i/o with and without the extender.

Without extender:

Sequential write speed 21586 KB/sec (target 10000) - PASS
Random write speed 704 IOPS (target 500) - PASS
Random read speed 2494 IOPS (target 1500) - PASS
3 Likes
6

You are really thorough! Awesome!

3 Likes
7

He is absolutely thorough, perhaps to a fault, but we still love him anyway. :wink:

2 Likes
8

Possibly, though it depends on the design of the interface, the kind of SSD being used and its architecture, along with a whole host of other factors like the way the USB or other interface drivers are implemented.

It has more aspects than a Siamese cat has attitude! :wink:  So a superficial study like that is only valid within the confines of a specific environment and Pi and I would hesitate to generalize it.

Your own study contradicted the premise of that article.

What that test IS good for is validating individual system performance using a specific set of hardware to see if it works for you in particular.

2 Likes
9

Note:
There is no such thing as a “mostly read-only robot”, especially if it’s using a Linux spin on a Pi.

I always assume that any modern operating system more complex than a micro:bit or an Arduino sketch is a high write-pressure environment.

Perhaps the Pi respins are optimized to reduce write pressure, but it’s still not trivial.

1 Like
10

You’re only just seeing the tip of the iceberg :slight_smile:
/K

2 Likes