aws: add i7ie instance type

[syuu1228]

Adding preset io parameters of i7ie to scylla_cloud_io_setup, and also added i7ie to supported instance type on aws_instance class.

Closes #559

[yaronkaikov]

@syuu1228 while you doing that, can you consider also implementing it as a dictionary as described - #544 ?

[avikivity]

Where do these numbers come from?

Please add to patch changelog.

[syuu1228]

Unlike other clouds like Azure, EC2 does not have IOPS and bandwidth information on their document.
So all preset values are measured by iotune on target instances instead of referenced from the document.
I measured 3 times on each instance types and used average of the result.

Added same description on the changelog.

[avikivity]

ok

[roydahan]

previously @xemul had some cavets on how we calculate the numbers for i4i.
I want to make sure we're doing it transparently this time.
Let's have it approved from @xemul as well.

[syuu1228]

@syuu1228 while you doing that, can you consider also implementing it as a dictionary as described - #544 ?

I'm working on #544, but I found that it takes few more time to complete implementation so I send this first.
We can import i7ie io parameters to #544 patch later.

[yaronkaikov]

@syuu1228 please rebase

[syuu1228]

@yaronkaikov Rebased with latest master.

[yaronkaikov]

@avikivity Ok from your POV?

[avikivity]

yes

[roydahan]

Marking as blocked until it gets review for iops and BW numbers.

[roydahan]

@syuu1228 please add the raw output of perftune to the commit message.

Also, can you please elaborate on how you ran iotune?
Was it on our AMI? raw instance of i7ie? without installing scylla and starting it first?
(I'm asking becuase with i4i we noticed differences)

[syuu1228]

@syuu1228 please add the raw output of perftune to the commit message.

Added.

Also, can you please elaborate on how you ran iotune? Was it on our AMI? raw instance of i7ie? without installing scylla and starting it first? (I'm asking becuase with i4i we noticed differences)

Here's measurement environment details:

• Measured on i7ie.* instances with latest version of Ubuntu 24.04 LTS AMI
  (We cannot use Scylla AMI since we do want to measure single drive performance)
• Measured single local SSD w/o RAID0, since we simulate RAID0 performance on
  scylla_cloud_io_setup script from single drive performance
• Use iotune for the measurement, executed 3 times for each instance size
  and used average of the results
• Automated measurement by script: https://github.com/syuu1228/ec2_run_script

[roydahan]

Thanks for that.
Do we get the same numbers when we run it on Scylla AMI with raid0 on top of all disks?

[syuu1228]

Thanks for that. Do we get the same numbers when we run it on Scylla AMI with raid0 on top of all disks?

These are theoretical values, so the actual performance may be lower than these figures depending on CPU performance and other factors.

Since EC2 has a fixed number of Local SSDs, it is possible to measure the performance on the RAID directly and apply the actual performance to the parameters, but this would require re-measuring a significant number of instance types.

[roydahan]

Since EC2 has a fixed number of Local SSDs, it is possible to measure the performance on the RAID directly and apply the actual performance to the parameters, but this would require re-measuring a significant number of instance types.

I'm not asking to re-measure on all instances, but take 2-3 instance type and measure with iotune on the raid0 with Scylla AMI and provide results.
They don't have to be exact but at least in the same area.