· by Aldrin Montana
What problem does the paper solve? Is is important?
This paper tries to address the state of some distributed systems work, describe common traps, and identify key issues that must be addressed for distributed systems research to make promising advances.
How does it solve the problem?
This paper is primarily a discussion of problems that the authors have identified in approaches in distributed systems research.
What alternative approaches are there? Are they adequately discussed in the reading?
Alternate approaches span object-based approaches, low level mechanisms that properly distinguish between local and remote computation, and previous work in RPC (remote procedure call), shared memory systems, and protocols such as NFS.
How does this work relate to other research, whether covered in this course or not?
This work is related to distributed systems overall, particularly in: programming languages (how are pointers/references presented), distributed protocols (a critique of the NFS protocol), and I think has interesting relevance to multi-core systems which I do not think existed at the time (in 1994, Intel had released the P54 processor, with 100 MHz and 166.3 MIPS), and so has interesting caveats that the authors did not accurately predict.
What specific research questions, if any, does the paper raise for you?
How does the proposed distinction between local and remote memory access apply for distributed object or key-value stores where data is logically accessed by key or entity and low-level access is managed by a server node? Does this fit "cleanly" into the described necessity to provide programmers with an appropriate interface so that they know what is local and what is remote? Also, for multi-core systems such as the Silo DBMS, does the distinction between local and remote need a third location? How important is it to create a third distinction, or distinguish not between local and remote but between processors separated by various scales of communication latency?