Discussion on Changes to netkvm RX Capacity Default Value

[zjmletang]

I’ve noticed that the default value for netkvm RX capacity has been changed from 256 to 1024. The reason provided is to make the configuration more user-friendly, as seen in RHEL-19430. My understanding is that this effectively hands over the decision of the default value to the backend. I would like to discuss a few points regarding this change:

1. This change may introduce an issue. If a cloud vendor believes that the most suitable default value for Windows is still 256 (which is quite likely, considering that netkvm still consumes a significant amount of memory), it seems that the only option would be to pass in 256 during the instance creation in QEMU. Once the instance is created, users would not be able to increase the RX capacity beyond 256 within the guest due to the limitation posed by the ring size of 256.

2. Additionally, there might be a minor side effect for cloud vendors who may prefer not to differentiate between Windows and Linux in QEMU. They might choose a unified larger value (such as 1024 or 4096). However, for a larger ring size like 1024, using it directly on Windows could lead to excessive memory usage, making it less suitable as a default value.

3. Lastly, regardless of the backend configuration options or adjustments within the guest, what would be the most appropriate value for the current netkvm? Personally, I still believe that 256 is the optimal choice.