Messages in this thread | | | From | takeru hayasaka <> | Date | Wed, 11 Oct 2023 14:25:55 +0900 | Subject | Re: [PATCH net-next] ethtool: ice: Support for RSS settings to GTP from ethtool |
| |
Thank you for teaching me so much!
> > Sorry! My email was blocked because it wasn't sent in plain text mode. > > I've made the necessary changes and will resend it. I got it. I'll do it like that :) Please let me know if I'm using it incorrectly.
> Makes sense, thanks for the extra information. I think it would be > worth adding all of this to the commit message!
of course:) I got it. I will add this background information when I issue a patch for the next version.
> Regarding the patch - you are only adding flow types, not a new field > (which are defined as RXH_*). If we want to hash on an extra field, > I think we need to specify that field as well?
I've been really struggling with this... When I read the Intel ICE documentation, it suggests that in RSS, TEID can be an additional input. However, I couldn't think of a reason not to include TEID when enabling RSS for GTP cases.
https://www.intel.com/content/www/us/en/content-details/617015/intel-ethernet-controller-e810-dynamic-device-personalization-ddp-technology-guide.html (cf. Table 8. Patterns and Input Sets for iavf RSS)
However, for Flow Director, it's clear that you'd want to include the TEID field. But since I found that someone from Intel has already configured it to use TEID with Flow Director, I thought maybe we don't need to add the TEID parameter for now.
https://patchwork.ozlabs.org/project/intel-wired-lan/cover/20210126065206.137422-1-haiyue.wang@intel.com/
If we want to include something other than TEID (e.g., QFI) in Flow Director, I think it would be better to prepare a new field.
2023年10月11日(水) 11:10 Jakub Kicinski <kuba@kernel.org>: > > On Wed, 11 Oct 2023 10:56:17 +0900 takeru hayasaka wrote: > > GTP generates a flow that includes an ID called TEID to identify the > > tunnel. This tunnel is created for each UE (User Equipment). > > By performing RSS based on this flow, it is possible to apply RSS for > > each communication unit from the UE. > > Without this, RSS would only be effective within the range of IP addresses. > > For instance, the PGW can only perform RSS within the IP range of the SGW. > > What I'm trying to say is that RSS based solely on IP addresses can be > > problematic from a load distribution perspective, especially if > > there's a bias in the terminals connected to a particular base > > station. > > As a reference that discusses a similar topic, please see the link > > below(is not RSS, is TEID Flow): > > https://docs.nvidia.com/networking-ethernet-software/cumulus-linux-56/Layer-3/Routing/Equal-Cost-Multipath-Load-Sharing/#gtp-hashing > > Makes sense, thanks for the extra information. I think it would be > worth adding all of this to the commit message! > > Regarding the patch - you are only adding flow types, not a new field > (which are defined as RXH_*). If we want to hash on an extra field, > I think we need to specify that field as well? > > > Thank you for your understanding. > > --- > > Sorry! My email was blocked because it wasn't sent in plain text mode. > > I've made the necessary changes and will resend it. > > No worries! Additional request - in the future please prefer the > bottom-posting or interleaved style of replies: > https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
| |