Markdown Version | Session Recording

Session Date/Time: 27 Mar 2023 08:30

alto

Summary

The ALTO working group meeting focused on deployment experiences and updates on key working items. Discussions covered cascading ALTO deployments, integration with the LHC Open Network Environment (LHC OPN), Tele Cdn pilots, and proposals for multi-domain and compute exposure extensions to ALTO. The meeting highlighted the challenges and potential solutions for deploying ALTO in complex, multi-domain networks, and explored new use cases for ALTO in service function chaining and compute resource optimization.

Key Discussion Points

• Cascading ALTO Deployments (Igor):

  • Discussed the challenges of ALTO adoption, noting the "chicken and egg" problem where networks and hyper giants are waiting for the other to adopt ALTO first.
  • Highlighted the potential for multi-domain ALTO to benefit ISPs even without hyper giant adoption.
  • Emphasized the difficulty of determining ingress points in networks for accurate path building.
  • Presented a use case where ALTO could have mitigated traffic issues during an iOS update by providing prior mapping and capacity information.

• Integration with LHC OPN (Richard):

  • Overviewed the LHC OPN environment and its use cases for supporting high-energy physics experiments.
  • Described the application-defined networking architecture and how ALTO is being integrated for transport scheduling.
  • Explained how ALTO provides visibility into network resource utilization and helps optimize data transfers.
  • Presented ALTO as a unified interface for sorting replica locations for data downloads.

• Tele Cdn Pilot (Luis):

  • Provided an update on the Tele Cdn pilot, highlighting the goal of automating traffic delivery decisions based on real-time network information.
  • Discussed challenges related to router operating system updates, TCP window flapping from route reflectors, and efficient processing of data received via the xBGP speaker.
  • Outlined next steps including analyzing network dynamics, resolving OS dependencies, and continuing the qualification of ALTO.
  • Raised concerns about securing ALTO interactions and information, especially when exposing ALTO to external applications.

• Multi-Domain ALTO (Jody):

  • Explained the motivation for multi-domain ALTO due to the increasingly complex and interconnected nature of networks.
  • Discussed the challenges of path reconstruction and cost measurement in multi-domain environments.
  • Presented examples of multi-domain networks, including science networks and networks with bandwidth constraints.
  • Outlined two approaches for multi-domain ALTO: horizontal information exchange between ASes and a hierarchical approach with an aggregate ALTO server.
  • Addressed the issue of incremental deployment and provided guidance for scenarios where some ASes do not participate in the multi-domain ALTO framework.

• Compute Exposure (Luis, Jody):

  • Introduced two drafts: one on service awareness and the other on service function chaining.
  • Explained the rationale for integrating compute information with ALTO to optimize service placement and function chaining.
  • Highlighted the importance of compute exposure in the context of edge computing and distributed AI training.
  • Proposed combining network information with compute metrics to assist in identifying proper data centers and service instances.

Decisions and Action Items

• Action Item: People interested in the drafts should review each draft and give comments.
• Action Item: Luis to summarize the improvements/issues from Tele Cdn pilot, document it in the wiki and bring it to one of the interim meetings.

Next Steps

• The working group will hold a series of interim meetings to further discuss the multi-domain and compute exposure proposals.
• The discussions should involve operators to gather real-world deployment experiences.
• The working group will continue to work on the Tele Cdn pilot and report progress at the next IETF meeting.