File size: 1,955 Bytes
385bde5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
---
license: cc-by-nc-4.0
task_categories:
  - other
language:
  - en
tags:
  - cxl-ssd
  - recommendation-system
  - kvcache
  - mqsim
  - dlrm
  - paper-archive
  - cxlssd-archetype-routing
size_categories:
  - 10G<n<100G
pretty_name: "CXL-SSD Archetype Routing — Generated Traces"
---

# CXL-SSD Archetype Routing — Generated Traces

Stage-00 snapshot (2026-05-02) of processed access traces used in the
**CXL-SSD page-oriented embedding lookup** paper.

These are MQSim/MaxEmbed-format traces derived from public recommendation
datasets (MovieLens, Criteo, Avazu, Taobao, Amazon, Yelp, …) and the Qwen
KV-Cache traces. They feed into the Cylon (FEMU) and MQSim simulators.

## Companion repos

| Component | URL |
|---|---|
| Outer paper repo | https://github.com/shadowcollecter/cxlssd-archetype-routing |
| Cylon-full (FEMU CXL-SSD) | https://github.com/shadowcollecter/cylon-full-paper |
| MQSim-CXL fork | https://github.com/shadowcollecter/mqsim-cxl-paper |
| MaxEmbed | https://github.com/shadowcollecter/maxembed-paper |

## Layout

Each subdirectory corresponds to one source dataset and contains traces
of the form `<layout>_<ratio>.trace` plus the matching MQSim
`workload.xml` / config snippets.

The trace ASCII format (per MQSim convention):

```
<arrival_time_ns> <device_num=0> <LSN_byte_addr> <size_sectors=8> <type 0=W|1=R>
```

Page-oriented variant: LSN is the byte address of a 4 KB page; size = 8
sectors (4 KB).

## Provenance

Generated from `/research_data/raw/<dataset>/` via the pipelines in
`MERCI_page_aware/analysis/`, `trace_converter/`, and
`MaxEmbed/scripts/`. See `MANIFEST.md` for per-dataset preprocessing
parameters.

## License

Trace files derived from each upstream dataset inherit the upstream
licence. This collection (the bundle layout + associated metadata) is
released **CC-BY-NC-4.0** for academic reproducibility.

Do not redistribute Criteo/Avazu/Taobao/Amazon raw fields outside
academic context.