====== BORG ====== ===== FAST 09, Camera Ready Paper ===== * 07/21/09: BORG on Argonne storage system discussion {{{{internal:projects:borg:medha072109b.mp3|mp3}} * 07/21/09: Medha defense practice talk {{{{internal:projects:borg:medha072109a.mp3|mp3}} * 07/02/09: Medha Thesis Intro chapter discussion {{{{internal:projects:borg:medha070209.mp3|mp3}} * 02/19/09: Medha FAST practice talk {{{{internal:projects:borg:borg021909.mp3|mp3}} * 12/16/08: Camera-ready version discussion {{{{internal:projects:borg:borg121608.mp3|mp3}} * 11/13/08: Discussion of To-Do List {{{{internal:projects:borg:borg111208.mp3|mp3-part1}} {{{{internal:projects:borg:borg2-111208.mp3|mp3-part2}} ===== Open Questions ===== * Are we measuring the foreground overhead of reconfiguration? How? * How are we changing the graphs in Figure 1 to make them smaller? ===== Modifications ===== ===== TODO ===== * Format paper according to FAST specifications. * Re-do EPS to include fonts. * Modify Figure 1 so is more visible. * Correct Figure 2. * State why we use throughput instead of response time. * Explain why we use 168x as acceleration factor. * Explain why it is OK to run 1x while reconfiguration is running. * Add a section of discussion where we address reviewers suggestions. * Events that trigger reconfiguration. * State what "certain threshold" means in section 4.3. * Change the name of OPT to be BOPT. * Add: When the size of the map in memory is prohibited, the size of the OPT should be reduced. * References: * Mac hfs' on-the-fly defragmentation. * Lumb et al. and Thereska et al. * Explain the indirector behavior when the write region is full and a new write request is issued. * Explain what values we take from blktrace. * Simplify Figures 6, 7, 10 and 11 to make them more readable. * Reduce the space used by referencing EXCES when possible. * Add more detail of traces in section 2. * Introduction in section 5.3. * Sort references. ===== TODO Email (From Luis to Ken) ===== Dear Ken, Here is a list of changes that we intend to make for the camera-ready version: - In the introduction, we will clarify the graphs in Figure 1 and possibly add additional graphs (space permitting) to clearly show details of workload characteristics. - In Section 4, we will clarify certain sections to address the following reviewer questions: * Explain what values of trace data do we collect from blktrace? * What does "certain threshold" mean in section 4.3? * How does the indirector behave when the write region is full and a new write request is issued? - We will explain the relevance of section 5.3 to BORG. - In the evaluation section we will justify: * Why we chose throughput as our metric versus response time. * Why we use 168X as acceleration factor in the non-reconfiguration phases and 1X in the reconfiguration phase. * Simplify Figures 6, 7, 10 and 11 to make them more readable. - Rename the acronym "OPT" to "BOPT" (for BORG OPtimized Target partition -- other suggestions are very welcome!) - Add the missing references: * Mac hfs' on-the-fly defragmentation. * Lumb et al. and Thereska et al. work on freeblock scheduling * EXCES, our previous work on external-caching - Certain reviewer comments require more discussion. We plan to address these in an additional section titled "Discussion" to appear before the Conclusion section: * What events should trigger reconfiguration? * Alternative layout algorithms for the BOPT? * Is there a way for BORG to degrade performance gracefully? - Format the paper based on the FAST specifications - Address the typos, sort references Please let us know if you have additional suggestions for improving the camera-ready version. We will address these and (other suggested changes from you, if any) and send you a version within the next 3-4 weeks. If there is anything else you need from us, please let me know. I will be the contact author for this paper. Luis' signature