Proceedings of the 7th symposium on Operating systems design and implementation 论文

详细信息

发表日期

2006-11-06

发表年份

2006

摘要

We are very happy to welcome you to Seattle for the 7th USENIX OSDI conference. Nearly twelve years ago, OSDI was conceived as the off-year SOSP. OSDI has always tried to attract and publish papers of the same depth and quality as those found in SOSP, but with somewhat more emphasis on implementation. We would like to think that OSDI is no longer a fill-in between SOSPs, but a true peer. You (and time) will tell whether OSDI has achieved that, but here we will say a little about how we ran the OSDI '06 process in support of this goal. OSDI '06 received 155 submissions. The Call for Papers gave specific instructions for paper length and format. We asked six authors to withdraw their papers due to substantial non-compliance with these requirements, out of fairness to those authors who did follow the rules. The remaining papers were reviewed in two rounds. In the first round, each paper received full reviews by three PC members, and many were also reviewed by an external reviewer. Based on the first-round reviews, the PC promoted 87 papers to the second round. Second-round papers received at least two additional full reviews by PC members, and many received additional PC or external reviews. This meant that every paper discussed at the PC meeting had been reviewed by at least five PC members. In all, we generated 697 reviews, most of them quite detailed. We accepted 27 of the 149 submissions, about 18%; this is in line with the historical average for OSDI, and consistent with the top conferences in the field. As traditional for OSDI, we assigned a PC member to shepherd each accepted paper. We allowed PC members to submit papers, but we judged these very carefully. Of the five papers submitted by PC members, we accepted just one. As program chairs, we were sensitive to the accusation that systems conferences often default to accepting papers that are clearly correct, yet boring; this leads to conferences short in novelty, and controversy. Consequently, we nudged the OSDI reviewers to favor but flawed submissions over but correct ones, as long as the flaws were minor and likely fixable. Reviewers assigned scores on a scale of 1--5 for quality, excitement, and quality. After the program was set, we created scatter plots showing the relationship between these metrics and the final outcomes. Figure 1 shows that boring papers tended to not make it past the first round, but that exciting papers, even imperfect ones, tended to end up in the conference. Nevertheless, we ultimately rejected some highexcitement papers because we were not satisfied with their technical quality. Presentation quality also turned out (not surprisingly) to be an excellent predictor of acceptance. Most of the papers that were highly ranked in terms of presentation were ultimately accepted as shown in Figure 2. Good writing improves the reviewers' understanding of the technical content, and makes it easier to appreciate the authors' own excitement.