shadowcollecter commited on
Commit
385bde5
·
verified ·
1 Parent(s): 0665a59

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +67 -0
README.md ADDED
@@ -0,0 +1,67 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: cc-by-nc-4.0
3
+ task_categories:
4
+ - other
5
+ language:
6
+ - en
7
+ tags:
8
+ - cxl-ssd
9
+ - recommendation-system
10
+ - kvcache
11
+ - mqsim
12
+ - dlrm
13
+ - paper-archive
14
+ - cxlssd-archetype-routing
15
+ size_categories:
16
+ - 10G<n<100G
17
+ pretty_name: "CXL-SSD Archetype Routing — Generated Traces"
18
+ ---
19
+
20
+ # CXL-SSD Archetype Routing — Generated Traces
21
+
22
+ Stage-00 snapshot (2026-05-02) of processed access traces used in the
23
+ **CXL-SSD page-oriented embedding lookup** paper.
24
+
25
+ These are MQSim/MaxEmbed-format traces derived from public recommendation
26
+ datasets (MovieLens, Criteo, Avazu, Taobao, Amazon, Yelp, …) and the Qwen
27
+ KV-Cache traces. They feed into the Cylon (FEMU) and MQSim simulators.
28
+
29
+ ## Companion repos
30
+
31
+ | Component | URL |
32
+ |---|---|
33
+ | Outer paper repo | https://github.com/shadowcollecter/cxlssd-archetype-routing |
34
+ | Cylon-full (FEMU CXL-SSD) | https://github.com/shadowcollecter/cylon-full-paper |
35
+ | MQSim-CXL fork | https://github.com/shadowcollecter/mqsim-cxl-paper |
36
+ | MaxEmbed | https://github.com/shadowcollecter/maxembed-paper |
37
+
38
+ ## Layout
39
+
40
+ Each subdirectory corresponds to one source dataset and contains traces
41
+ of the form `<layout>_<ratio>.trace` plus the matching MQSim
42
+ `workload.xml` / config snippets.
43
+
44
+ The trace ASCII format (per MQSim convention):
45
+
46
+ ```
47
+ <arrival_time_ns> <device_num=0> <LSN_byte_addr> <size_sectors=8> <type 0=W|1=R>
48
+ ```
49
+
50
+ Page-oriented variant: LSN is the byte address of a 4 KB page; size = 8
51
+ sectors (4 KB).
52
+
53
+ ## Provenance
54
+
55
+ Generated from `/research_data/raw/<dataset>/` via the pipelines in
56
+ `MERCI_page_aware/analysis/`, `trace_converter/`, and
57
+ `MaxEmbed/scripts/`. See `MANIFEST.md` for per-dataset preprocessing
58
+ parameters.
59
+
60
+ ## License
61
+
62
+ Trace files derived from each upstream dataset inherit the upstream
63
+ licence. This collection (the bundle layout + associated metadata) is
64
+ released **CC-BY-NC-4.0** for academic reproducibility.
65
+
66
+ Do not redistribute Criteo/Avazu/Taobao/Amazon raw fields outside
67
+ academic context.