-
Here’s our scenario. We have tart VMs running on AWS EC2 mac2.metal instances for our iOS build pipelines. Tart is a great tool, thank you!! However the developers are complaining of slow CI build times. For example a build on the developer’s laptop takes 10 minutes, but on the EC2 instances it takes > 20 minutes. I’ve think I’ve narrowed the problem down to slow write speed but I’m stuck trying to work out why. I’m using a simple dd to push the throughput as hard as I can. But I see the same kind of thing running the fio tool that Amazon recommend. On the host mac I’m getting good write performance on gp3 EBS. If I configure 400 MB/s throughput I can push 400 MB/s writes from the host. No problem. Running a tart VM (both host and guest are Sonoma) with the disk on the host’s EBS volume I get terrible write throughput < 50 MB/s. If I run the Tart VM with its disk on the internal SSD I get much faster throughput > 500 MB/s. Something about the EBS volume and the Virtualisation layer is not playing nice. Any clues? I’ve tried maxxing out the IOPS, but that doesn’t help. I've tried
|
Beta Was this translation helpful? Give feedback.
Replies: 3 comments
-
Hello David 👋 A couple of questions:
|
Beta Was this translation helpful? Give feedback.
-
It worked!!! So happy. You are a legend. My test has been as simple as On my 400MB/s EBS volume I'm now seeing the full 400MB/s being used. This is a total game changer. Thank you. The devs will be so happy their builds are going to fly along. |
Beta Was this translation helpful? Give feedback.
-
thank you tart team, this is amazing, i had the same issue, with new release and the arg, the performance is improved. |
Beta Was this translation helpful? Give feedback.
Hello David 👋
A couple of questions:
How are you measuring the throughput?
We've just released Tart
2.21.0
, which now allows you to explicitly request data caching for the disk I/O.Would you mind repeating the experiment with
--root-disk-opts="sync=none,caching=cached"
?