this is for holding javascript data
Brandon Holt add Lynx/transaction chains
over 9 years ago
Commit id: b952edb2f7b5f90dcdc9ac95c06b160da3f885e8
deletions | additions
diff --git a/Questions.md b/Questions.md
index e0d114f..0936ec3 100644
--- a/Questions.md
+++ b/Questions.md
...
- in Spark?
- in an existing database?
- Redis? ("data structures server", but not designed to be consistent across replicas)
- would people take this seriously if it was a modification to Redis?
- Tapir?
- Doppel? (Phase Reconciliation paper)
- code not (yet) available
- Lynx / Rococo
- Distributed key/value store
- Existing DTM framework
- HyFlow (Scala)
- how best to evaluate?
diff --git a/Related work.md b/Related work.md
index 7a3f42a..4a7bc9e 100644
--- a/Related work.md
+++ b/Related work.md
...
- doesn't do replication for hot keys
- how pieces are reordered seems a bit crude (anything that produces a value is "immediate" and cannot be deferred)
- seems like it would have high runtime overhead (tracking dependences dynamically, despite using offline (static?) analysis)
- Citations
- ### Transaction chains
- Prior work \cite{Zhang:SOSP13} (same
authors): *Lynx*; statically split authors)
- *Lynx*: geo-replicated key/value store
- programmer splits transactions into pieces that run on different
locations, either the locations
- whole chain
gets committed or none is executed in a serializable way, can only *fail* at the first hop
- static analysis to determine if the above can be enforced
- *Benchmarks*
- Social: FB-like social graph, users, walls, activity feeds
- Twitter: Users, tweets, directed graph
- RUBiS
- *Questions*
- not thinking about high contention cases? does that come up in the social graphs?
## Enhancing Concurrency in DTM through Commutativity
- \cite{Kim:EuroPar13}: EuroPar'13, Junwhan Kim, Roberto Palmieri, Binoy Ravindran