Wiki

Status Report

Presentation of workload format for trace-driven simulations.
- Three (3) message types defined: Insert, Update, Lookup
- Insert: A new binding is being added for the GUID value. Any existing values are preserved.
- Update: An old binding is replaced for the GUID value. The previous binding and new binding are required. A missing old value means the Update will be treated as an Insert.
- Lookup: All current (non-expired) bindings for a GUID value are returned.
K-replica code works now on multiple servers
- Merged with cache code, but needs to be tested at runtime with traces
Trace-based simulations will use a file format proposed by Feixiong and agreed upon in the meeting
- The format will be maintained on the Wiki.
- "Client ID" will replace "Source AS", "Time Offset" will replace "Interval".

How can we make GNRS ready for SIGCOM publication?
Should a Delete message be included? What are the semantics involved?
- How will policy be enforced with multi-homed/sourced content GUIDs?
- Need to ensure a discussion is included in any performance paper, as Delete may have performance impacts.
When GUID mappings expire within a server, we should ensure that any clean-up related tasks do not impact response times.
- Make sure the response message is generated BEFORE any clean-up tasks are handled by the thread afterward.
For a Lookup message, we should consider how to handle limiting the number of returned quantities.
- How is the limit specified? Which bindings are returned to the client?
Discuss what accurate models/heuristics exist for different mobility/access characteristics.
Different "domains" of access
- 1. Human communication - e.g., calls, SMS, email
- 2. "Web" content/browsing - access to web servers/SaaS
- 3. "M2M" control - push/pull background data updates. Programmatically-generated.
- Paper Idea: Sniff a bunch of iPhone/Android traffic and try to characterize/classify based on broad categories above.

Determine how to generate traces for SIGCOM: read SIGCOM papers.
- Mobile IP (example of a failed system)
Test integrated code with K-replica and caching
Evaluate performance of multiple replicas (1-5), also look at effect of caching, cache size, ttl
Write a script to deploy and run an experiment using some simple topology/trace.