deletions | additions       diff --git a/Related work.md b/Related work.md  index ed5240e..b585a4b 100644  --- a/Related work.md  +++ b/Related work.md 
...
- what happens when *part* of a transaction has to wait until a joined phase (if run during a split phase)? Does the transaction abort? Suspend and resume and commit later?  - What is the performance like if you have an `incr` and `read` in the *same* transaction, on the same record? Didn't see if there were any performance numbers for that, but seems like it would probably interact badly with the phasing.  - escrow transactions (Neha got a question about these after her talk)  - Citations  - RedBlue consistency \cite{Li:OSDI12}: separate ops into "blue" fast ops which don't require strict consistency, and "red" which are slower but more general. uses same RUBiS benchmark  - Conflict-free replicated data types \cite{Shapiro:SSS11}: will converge under eventually-consistent conditions  ## Extracting More Concurrency from Distributed Transactions  - \cite{Mu:OSDI14} Rococo - two-phase protocol, implemented on distributed transactional db 
...
- 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 \cite{Zhang:SOSP13} (same authors): *Lynx*; statically split transactions into pieces that run on different locations, either the whole chain gets committed or none  ## Enhancing Concurrency in DTM through Commutativity   - \cite{Kim:EuroPar13}: EuroPar'13, Junwhan Kim, Roberto Palmieri, Binoy Ravindran