brycewang-stanford ASPLOS Related Work
An ASPLOS reviewer pool is assembled from three communities, and each reviewer checks whether their literature is handled competently. Related work at this venue is therefore a coverage problem across parallel lanes — a paper positioned beautifully against the OS literature and blankly against the architecture literature reads as half-researched to half its reviewers.
The five lanes
| Lane | What must be shown | Failure the lane's reviewers spot |
|---|---|---|
| Architecture (ISCA/MICRO/HPCA lineage) | The hardware structure or mechanism is new relative to prior microarchitectural work | Reinvented prefetcher/hint/metadata scheme under a new name |
| OS/systems (SOSP/OSDI/EuroSys/ATC lineage) | The policy/runtime layer differs from existing kernel or hypervisor mechanisms | "The kernel already does this with tunables" |
| PL/compilers (PLDI/POPL/CGO lineage) | Language/compiler-level alternatives are acknowledged when the problem admits them | Hardware solution to a problem the compiler community solved statically |
| Accelerator/domain (when applicable) | Placement in the accelerator design space with honest technology normalization | Comparing against a three-generation-old design point |
| Prior ASPLOS | The intersection lineage itself — what earlier cross-layer attempts did | Missing the exact ASPLOS paper a PC member wrote or shepherded |
For the last lane, mine the venue's own memory: the ACM DL ASPLOS series and the
Influential Paper Award list (see ../../resources/exemplars/library.md) identify
the papers the community regards as canonical for a given boundary.
Differentiate on the boundary, not the components
The distinctive ASPLOS positioning move: prior work usually shares components with yours (a similar hardware structure, a similar kernel policy) while drawing the hardware/software boundary elsewhere. Say precisely where each prior system draws it and why yours moves it:
Template, per closest-prior system:
"<System> provides <capability> at <layer>, which requires <cost/assumption>.
Because it keeps <decision> in <layer>, it cannot <the thing your boundary
placement enables>. We move <decision> to <layer>, which is what makes
<headline property> possible."
Three such paragraphs beat fifteen one-line dismissals — and the systems that get this treatment should be the ones a hostile reviewer would name.
Double-blind handling of your own line of work (2027 rules)
- Cite your prior published work in the third person, treated exactly like a competitor's system.
- Your own arXiv postings and tech reports are not prior publications under the 2027 CFP, and the submission may ignore them — you are not forced into contortions differentiating against your own preprint.
- A workshop paper this submission extends, or your related manuscript under review
elsewhere, goes in as anonymized supplemental material so reviewers can judge
the delta (mechanics in
asplos-supplementary). - Never use "removed for blind review" placeholders — disallowed wording.
Format constraints that reshape this section
ASPLOS's citation rules (verified 2026-07-08) have an underappreciated interaction
with related work: every entry needs all co-authors' full first and last names
and a DOI/link, which makes each citation long. Since references are outside the
11-page limit this costs no body pages — but it means bibliography hygiene is a
real task: budget an editing pass to expand names (asplos-submission has the
mechanical sweep) rather than discovering 90 malformed entries at the deadline.
Timing and staleness
- Sweep the most recent two editions of ASPLOS, ISCA, MICRO, SOSP/OSDI, and PLDI before each gate (the September 9, 2026 deadline follows a spring conference season — its output is your reviewers' freshest context).
- Concurrent work that appears between cycles: acknowledge it neutrally in a revision or camera-ready; a Major Revision's change note is an expected place to add "since submission" positioning.
Building the lane map before writing
Construct the section from a spreadsheet, not from memory:
- For each lane, list the 5-10 systems a reviewer from that community would expect to see; source them from the last two editions' proceedings tables of contents (ACM DL for ASPLOS/ISCA/MICRO/SOSP; the respective sites for USENIX and SIGPLAN venues) rather than from citation chains, which propagate one group's blind spots.
- Mark each entry: boundary-competitor (gets a template paragraph), component-overlap (grouped sentence), or context (grouped citation).
- Have the co-author closest to each community audit their lane's list — the embarrassing omission is almost always in the lane nobody on the team calls home.
- Recheck the map at every gate: between April and September a spring conference season publishes, and a reviewer's own new paper is the citation they miss first.
Using the venue's own canon
The Influential Paper Award list is a positioning instrument, not just an
exemplar source: if your boundary has an award-validated ancestor (an
accelerator argued from data movement, a virtualization boundary redraw — see
../../resources/exemplars/library.md), citing the lineage and stating your
delta against it signals venue literacy to every senior reviewer. Conversely,
discovering that a canon paper already occupies your claimed boundary is
cheaper during this pass than in a review.
Where the section lives
With 11 figure-inclusive pages, a late monolithic related-work section is expensive real estate. The pattern that survives compression: inline differentiation at first mention for the two or three boundary-competitors (they are part of the argument, not an appendix to it), plus a compact lane-organized section that proves coverage. What must not be compressed away is the boundary-competitor analysis — cutting it saves half a page and costs the review that says "does not engage with ."
Output format
[Lane coverage] arch / OS / PL / accelerator / prior-ASPLOS — each: covered / thin / n-a
[Boundary paragraphs] closest-prior systems given the template treatment: list
[Hostile-reviewer test] the 3 systems a skeptic would name — all addressed? Y/N
[Own-work handling] third-person OK · supplements needed · arXiv stance decided
[Bibliography hygiene] full-name + DOI pass scheduled/done
[Staleness] newest cited edition per lane