Re: [alto] New draft on joint exposure of network and compute information

[Jordi Ros Giralt <[e-mail]>]

Thanks Linda.

These metrics can be common to a variety of use cases. That is, the same common metrics can be used to support the CATS use cases (exposure to the ingress point) or (as suggested by this draft) to support service providers and applications to make service deployment and selection decisions. The two efforts (network and application layer) can complement each other to achieve the performance requirements.

As for metrics, it depends on the application, but certainly on the compute side, CPU, GPU, NPU, memory, storage are all important to make proper placement and selection decisions. A good part is that, in general, for a given application (e.g., AI inference), the requirements to run it are generally well-known at deployment/selection time.

Thanks,
Jordi
________________________________
From: Linda Dunbar <[e-mail]>
Sent: Wednesday, November 1, 2023 18:15
To: Jordi Ros Giralt <[e-mail]>; [e-mail] <[e-mail]>; [e-mail] <[e-mail]>
Subject: RE: New draft on joint exposure of network and compute information


WARNING: This email originated from outside of Qualcomm. Please be wary of any links or attachments, and do not enable macros.

Jordi,



In addition, CATS WG is discussing many more metrics that can impact the service performance.



Really appreciate if you can elaborate more on the new metrics that impact the deployment and path section. It will be very useful.



Thanks, Linda







From: Linda Dunbar
Sent: Wednesday, November 1, 2023 12:11 PM
To: Jordi Ros Giralt <[e-mail]>; [e-mail]; [e-mail]
Cc: [e-mail]
Subject: RE: New draft on joint exposure of network and compute information



Jordi,



Your draft describes two aspects of the service performance impacted by the Computing: Service Deployment and  Service (Path) Selection. Those two should be separated, as the Service Deployment belongs to the OpsArea, and the Service selection (including Network Path & DCs that host the services) belongs to the Routing area.



https://datatracker.ietf.org/doc/draft-ietf-idr-5g-edge-service-metadata/ has proposed a new Metadata Path Attribute and some Sub-TLVs  for egress routers to advertise the Metadata about the attached edge  services (ES).  The Edge Service Metadata can be used by the ingress routers in the 5G Local Data Network to make path selections not only based on the routing cost but also the running environment of the edge services.  The goal is to improve latency and performance for 5G  edge services.





Can this Metadata Path Attribute address the problem stated in your draft?  I CC’ed the IDR WG, so your comments on the Path Selection can be visible to them.



Thanks, Linda





From: Cats <[e-mail]<mailto:[e-mail]>> On Behalf Of Jordi Ros Giralt
Sent: Tuesday, October 24, 2023 6:47 AM
To: [e-mail]<mailto:[e-mail]>; [e-mail]<mailto:[e-mail]>
Subject: [Cats] New draft on joint exposure of network and compute information



Dear CATS and ALTO WG mailing list members,



We submitted a new draft on joint exposure of network and compute information for service placement and selection: https://datatracker.ietf.org/doc/draft-rcr-opsawg-operational-compute-metrics/




Joint Exposure of Network and Compute Information for Infrastructure-Aware Service DeploymentJoint Exposure of Network and Compute Information for Infrastructure-Aware Service Deployment

This draft focuses on the problem of exposing both network and compute information to the service provider/application to support service placement and selection decisions. ALTO provides an interface for network information exposure to the service provider/application; thus, an approach is to leverage and extend it with compute metrics. CATS also needs to develop compute metrics to support traffic steering decisions. The common ground is in these compute metrics, which could be reused across the various use cases (e.g., consumed by the network as in CATS or consumed by the application as in ALTO).



This draft also aims at providing a framework for continuing the discussion initiated during IETF 117 regarding the presentation "Compute-aware metrics: CATS working with ALTO": https://datatracker.ietf.org/doc/slides-117-alto-compute-aware-metrics-cats-working-with-alto/



We would like to seek feedback from both working groups on developing compute metrics that can be reused for different use cases, to avoid duplicated work and increase the effectiveness of future standards.



Thanks,

Jordi