Pith. sign in

REVIEW 5 major objections 4 minor 47 references

Efficient Vector Search on Disaggregated Memory with d-HNSW

T0 review · 5 major / 4 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read d-HNSW shows HNSW vector search can run over RDMA-disaggregated memory by routing each query through a small cached meta-graph and fetching only the relevant remote sub-graphs, cutting latency up to 117x at recall 0.87 on SIFT1M.

desk verdict The design is coherent but the 117x speedup is a measurement artifact; the naive baseline's RDMA latency is physically impossible, so the headline claim collapses. read the letter →

arxiv 2505.11783 v1 pith:U2T3RZ5R submitted 2025-05-17 cs.DB

Yi Liu , Fei Fang , Chen Qian This is my paper · ORCID
classification cs.DB
keywords disaggregatedmemoryRDMAvectorsimilaritysearchHNSWapproximatenearestneighborgraph-basedindexdoorbellbatchingpool
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

d-HNSW aims to make graph-based vector similarity search practical when the index and data live in a remote memory pool accessed over RDMA. The idea is to split HNSW into a tiny cached routing index (meta-HNSW) built from 500 sampled vectors and a set of sub-HNSW clusters stored remotely; each query routes through the cached meta-graph and then fetches only a few relevant clusters. A dedicated memory layout and batched, doorbell-coalesced RDMA reads cut the number of network round trips per query from several to near zero. The paper reports up to 117x lower latency than a naive remote HNSW while keeping recall around 0.87 on SIFT1M@1, which matters because vector datasets increasingly outgrow single-machine memory.

What carries the argument

The meta-HNSW: a three-layer HNSW built from 500 uniformly sampled vectors and cached in the compute pool. It acts as a coarse classifier: greedy routing through its layers ends at a bottom-layer vector that selects a sub-HNSW partition, and the query then searches only the nearest few such partitions held in remote memory. The mechanism that carries the argument is the observation that HNSW's greedy search visits a short, predictable path, so a cheap replica of the upper graph can decide which remote clusters to load, while an RDMA-friendly layout and doorbell-batched reads turn the rest of the search into a small number of bulk transfers.

What would settle it

Run d-HNSW on SIFT1M with the routing step in place but give the local sub-HNSW search a very large efSearch; if recall plateaus well below 1.0, the plateau measures the fraction of true neighbors that fall outside the clusters selected by the cached meta-graph. Alternatively, directly check, for a random sample of queries, whether the ground-truth top-1 neighbor lies inside the selected sub-HNSW clusters.

Watch

Extended reading notes

Core claim

d-HNSW claims that the greedy search path of HNSW can be exploited for memory disaggregation: instead of following pointer-chasing steps across the network, a query first traverses a small cached meta-HNSW to select the few sub-HNSW partitions likely to contain the true nearest neighbors, then performs one or a few bulk RDMA reads to load those partitions. The paper argues that this two-level lookup preserves accuracy because the meta-HNSW, built from a uniform sample, reliably classifies queries into the correct clusters. On top of this, the remote index is laid out as serialized cluster pairs with shared overflow space, and doorbell batching merges non-contiguous reads into single round trips. The reported result is up to 117x lower query latency than a naive remote HNSW at a recall of 0.87 on SIFT1M@1, with similar gains on GIST1M.

Load-bearing premise

The load-bearing premise is that a meta-HNSW built from just 500 sampled vectors routes each query to the sub-clusters that contain its true nearest neighbors; if that routing ever fails, no additional search effort inside the chosen clusters can recover the missed neighbors.

Editorial extensions

If this is right

  • Vector similarity search can be served from a memory pool with weak computational power, because compute nodes perform routing through the cached meta-graph and the memory node only answers bulk RDMA reads.
  • Query cost is set by the number of relevant sub-HNSW clusters fetched, not by total dataset size, so the approach scales to datasets that exceed any single node's memory.
  • Batched queries share RDMA transfers: a cluster requested by many queries is loaded once per batch, and recently used clusters are reused across batches.
  • Dynamic insertions are handled by overflow space shared between pairs of sub-HNSW clusters, so new vectors stay contiguous with their cluster and can be read in one transfer.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the 500-vector routing sample generalizes, the same meta-graph idea could support multi-level partitioning for billion-scale datasets, with memory pools holding nested sub-clusters.
  • The recall ceiling set by routing suggests an adaptive extension: monitor how often true nearest neighbors fall outside the selected clusters and rebuild or enlarge the meta-HNSW when needed.
  • The reported 117x gain is against a naive remote baseline; comparing against a monolithic in-memory HNSW on the same hardware would quantify the intrinsic cost of disaggregation, a comparison the paper does not include.
  • Shared overflow space implies a write-heavy workload could exhaust the reserved area; a testable extension is measuring insertion throughput and overflow behavior over sustained updates.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 4 minor

Summary. The paper introduces d-HNSW, a vector similarity search engine for RDMA-based disaggregated memory. The design partitions the dataset into sub-HNSW clusters, caches a small "meta-HNSW" in compute nodes for cluster routing, lays out clusters contiguously in remote memory with shared overflow regions, and uses doorbell batching plus query-aware data loading to reduce RDMA round trips. The evaluation reports latency-recall curves for SIFT1M and GIST1M against two self-defined baselines (Naive d-HNSW and d-HNSW without doorbell), claiming up to 117x lower latency while maintaining recall around 0.87 on SIFT1M@1.

Significance. Disaggregated memory for vector search is a timely and practically relevant problem, and the three proposed techniques (representative index caching, RDMA-friendly layout, and batched query-aware loading) are plausible building blocks. The paper also ships a substantial prototype implementation (12K LoC) and evaluates on real hardware, which is a strength. However, as submitted, the evaluation does not substantiate the central performance claims: the baselines are self-defined variants, the absolute recall figures are not benchmarked against standard HNSW, the network-latency numbers in the breakdown tables are physically implausible on the stated testbed, and the key routing assumption behind meta-HNSW is given no sensitivity analysis. If these issues are addressed, the work could be a useful contribution to the disaggregated-systems and vector-search communities.

major comments (5)
  1. [§4, Table 1] The Naive d-HNSW network latency of 90,271.2 µs for 3.547 RDMA round trips per query implies roughly 25 ms per RDMA read. On the stated Mellanox ConnectX-6 100Gb testbed, even a multi-megabyte RDMA read should complete in well under 1 ms; the value is most likely explained by a unit error, by measuring an entire batch of 2000 queries as one query, or by an uninstrumented serialization of many NIC operations. Because the headline 117x speedup is the ratio d-HNSW/Naive, an implausible denominator directly inflates the headline claim. The authors should provide a raw RDMA latency calibration on the testbed and re-measure the baseline before any performance claim can be accepted.
  2. [§4, Figure 6] The only comparisons are to two self-defined variants, "Naive d-HNSW" and "d-HNSW without doorbell." There is no comparison to standard in-memory HNSW, to a non-disaggregated ANN system, or to prior RDMA- or CXL-based vector-search prototypes despite such systems being cited in the paper. As a result, the absolute recall values (0.85–0.87 on SIFT1M@1 at efSearch=48) cannot be interpreted, and the claim in the abstract and introduction that d-HNSW is "the first" disaggregated vector search engine is unsupported. A fair evaluation needs at least a standard HNSW recall-latency curve on the same datasets.
  3. [§3.1, Figure 3] The correctness of the meta-HNSW routing layer is assumed rather than demonstrated. The paper states that the meta-HNSW is built from 500 uniformly sampled vectors and that each query is routed to the b closest sub-HNSW clusters, but it provides no analysis or sensitivity study showing that the selected clusters contain the true top-k neighbors. If the meta-HNSW misses the correct cluster, recall is capped regardless of the local search effort. The paper should report a sensitivity analysis over sample size and b, and should separately measure recall loss caused by routing versus recall loss caused by sub-HNSW search within the selected clusters.
  4. [§3.2, Table 1] The "round trips per vector query" metric is internally inconsistent with the reported network latencies. For d-HNSW the paper reports 0.00475 round trips per query but a network latency of 527.6 µs, which would imply roughly 111 ms per round trip; for the Naive baseline, 3.547 round trips and 90,271.2 µs imply roughly 25 ms per round trip. The relation between the doorbell batch size, the batch of 2000 queries, and the per-query latency aggregation needs a precise definition, otherwise the round-trip counts cannot be used as evidence for the speedup.
  5. [§4, experimental setup] Several experimental conditions that are load-bearing for the evaluation are missing: the number and size of sub-HNSW clusters, the distribution of vectors across clusters, the actual cache hit rates, the overhead of index construction and insertion, and run-to-run variance (no error bars are shown in Figure 6). The abstract claims "fast data indexing" and "low network communication overhead," but no indexing or insertion experiments are reported, so those claims are not evaluated.
minor comments (4)
  1. [§4, first paragraph] The baseline is introduced as "Native-HNSW" but is called "Naive d-HNSW" throughout the rest of the paper; the naming should be unified.
  2. [Abstract and §4] The abstract reports "up to 117x" speedup while Section 4 reports up to 121x on GIST1M; the inconsistency should be reconciled.
  3. [Figure 6] The latency-recall curves would be much more informative with error bars and with a labeled reference curve for a standard in-memory HNSW implementation.
  4. [Throughout] There are several typographical issues, including "evalute" in Section 4 and the formatting of "4.75×10^-3" for the round-trip count; these should be fixed in a revision.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the 117x result is an ablation against a deliberately non-batched baseline, not a prediction derived from its own definition.

full rationale

d-HNSW's core claims are empirical evaluations, not derivations. The meta-HNSW routing index and doorbell batching are design choices evaluated against Naive d-HNSW, which is explicitly defined as issuing one RDMA read per sub-HNSW cluster (Section 4). The resulting round-trip reduction (3.547 to 0.00475 per query, Table 1) is the mechanism under test, not a hidden equality between the input and the claimed result; a speedup over a deliberately non-batched baseline is an ablation, and the 'prediction' that batching reduces network latency would hold independently of the measured constants. No parameter is fitted on the target metric and then reported as a prediction. The paper's only self-citation (reference [16], Outback) appears in related work among other disaggregated KV stores and does not carry the argument. The plausibility concern about the absolute Naive network latency (90,271.2 µs for 3.547 round-trips in Table 1) is a measurement-validity issue, not a circularity: the latency number is not constructed from the speedup claim. Hence no circular step meets the quoted-evidence bar.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

The system relies on several hand-picked parameters (sample count, overflow size, cache ratio, batch size) and on the unproven routing accuracy of the meta-index. The empirical performance claim is therefore not an independent verification of the design.

free parameters (6)
  • meta-HNSW sample count = 500
    Number of vectors uniformly sampled to build the routing index; chosen without sensitivity analysis (Section 3.1).
  • overflow space per paired clusters = 0.75 MB (SIFT1M), 3.92 MB (GIST1M)
    Reserved for insertions in each group; values given per dataset with no derivation (Section 3.2).
  • cache capacity per compute instance = 10% of sub-HNSW clusters
    Cache size set to 10% of the total clusters; no sweep (Section 4).
  • query batch size = 2000
    Set at runtime; no sensitivity analysis (Section 4).
  • number of closest sub-HNSWs loaded per query (b)
    Paper does not report b, though it is central to the recall/latency tradeoff (Section 3.3).
  • doorbell batch size = 2 (example only)
    Used in the example in Figure 5; actual value not specified (Section 3.2).
assumptions (4)
  • standard math HNSW greedy search on a small-world graph returns near-optimal neighbors.
    Inherited from Malkov and Yashunin [20], used as the foundation of sub-HNSW search.
  • ad hoc to paper A meta-HNSW constructed from 500 uniformly sampled vectors partitions the data so that top-b selected clusters contain the true top-k neighbors with high probability.
    Core routing assumption; no proof, sensitivity analysis, or independent validation. Invoked in Section 3.1.
  • domain assumption Doorbell batching of multiple RDMA reads into one round trip reduces latency without exceeding NIC limits.
    Standard RDMA behavior, but the paper does not evaluate the tradeoff at scale; used in Section 3.2.
  • domain assumption Compute pool has limited DRAM and memory pool has weak CPUs.
    Defines the disaggregated scenario in Section 3, but no evidence that this configuration represents real deployments.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Efficient Vector Search on Disaggregated Memory with d-HNSW." pith.science (2026). https://pith.science/paper/U2T3RZ5R

@misc{pith2026250511783,
  author       = {Pith},
  title        = {Pith review of: Efficient Vector Search on Disaggregated Memory with d-HNSW},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/U2T3RZ5R}},
  note         = {Machine review of arXiv:2505.11783}
}
read the original abstract

Efficient vector query processing is critical to enable AI applications at scale. Recent solutions struggle with growing vector datasets that exceed single-machine memory capacity, forcing unnecessary data movement and resource underutilization in monolithic architectures. We present d-HNSW, the first disaggregated vector similarity search engine for RDMA-based remote memory systems that achieves high performance while supporting fast data indexing with low network communication overhead. The core of d-HNSW is a novel disaggregation of the graph-based vector indexing data structure HNSW. It exploits the characteristics of greedy searching in HNSW to efficiently coordinate data transfers from the memory pool to the compute pool while serving data requests. Specifically, it leverages three ideas: (i) Representative index caching, a lightweight index constructed from a sampled subset of data, is cached in the compute pool to reduce frequent access to critical components of the hierarchical graph-based index, (ii) RDMA-friendly data layout design to reduce the networking round trips incurred by vector query and insertion and (iii) batched query-aware data loading to reduce bandwidth usage on data transfer between pools, addressing the limited cache capacity in compute nodes. We evaluate d-HNSW with extensive benchmarking datasets. The experimental results show that d-HNSW outperforms Naive d-HNSW implementation by up to 117x in latency while maintaining recall as 0.87 in dataset SIFT1M@1.

Figures

Figures reproduced from arXiv: 2505.11783 by the authors.

Figure 1
Figure 1. Graph-based vector search index: HNSW. systems [10] by enabling fast, high-dimensional similarity searches over massive embedding spaces. In RAG, a vector database retrieves semantically relevant documents based on the user prompt’s embedding, allowing LLMs to generate responses with external knowledge rather than limited to the information encoded in their model parameters. As these AI-driven applications [8, 29] c… view at source ↗
Figure 2
Figure 2. The overview of d-HNSW. and LSH [7] struggle with scalability and search accuracy in high-dimensional spaces, leading to the development of graph-based indexing techniques [6, 20]. These methods con￾struct a navigable graph where data points serve as nodes, and edges encode proximity relationships, enabling fast tra￾versal during queries. For example, as shown in [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Representative index caching in d-HNSW. vector query processing. d-HNSW achieves so by represen￾tative index caching (§3.1), RDMA-friendly graph index stor￾age in remote memory (§3.2), and query-aware batched data loading (§3.3). Here, we provide a brief overview of d-HNSW as [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (2 more)
Figure 5
Figure 5. Figure 5: Query-aware sub-HNSW clusters loading. serialized HNSW index and floating-point vectors. At the beginning of this memory space, a global metadata block records the offsets of each sub-HNSW cluster, as their sizes vary. The remaining memory space is divided into groups,…
Figure 6
Figure 6. Figure 6: Latency-recall evaluation of d-HNSW and baselines. [PITH_FULL_IMAGE:figures/full_fig_p005_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

47 extracted references · 40 canonical work pages

  1. [1]

    Cowbird: Freeing cpus to compute by offloading the disaggregation of memory

    Xinyi Chen, Liangcheng Yu, Vincent Liu, and Qizhen Zhang. Cowbird: Freeing cpus to compute by offloading the disaggregation of memory. InProceedings of the ACM SIGCOMM 2023 Conference, pages 1060–1073, 2023

    work page 2023

  2. [2]

    CloudLab: Flexible

    scientific infrastructure for research on the future of cloud computing. CloudLab: Flexible. https://www.cloudlab.us

    work page

  3. [3]

    https://www.deepseek.com/

    DeepSeek. https://www.deepseek.com/

    work page

  4. [4]

    Pyramid: A general framework for distributed similarity search on large-scale datasets

    Shiyuan Deng, Xiao Yan, KW Ng Kelvin, Chenyu Jiang, and James Cheng. Pyramid: A general framework for distributed similarity search on large-scale datasets. In2019 IEEE International Conference on Big Data (Big Data), pages 1066–1071. IEEE, 2019

    work page 2019

  5. [5]

    https://github.com/deepseek-ai/3fs/

    Fire flyer file system. https://github.com/deepseek-ai/3fs/

    work page

  6. [6]

    Fast approximate nearest neighbor search with the navigating spreading-out graphs

    Cong Fu, Chao Xiang, Changxu Wang, and Deng Cai. Fast approximate nearest neighbor search with the navigating spreading-out graphs. PVLDB, 12(5):461 – 474, 2019

    work page 2019

  7. [7]

    Similarity search in high dimensions via hashing

    Aristides Gionis, Piotr Indyk, Rajeev Motwani, et al. Similarity search in high dimensions via hashing. InVldb, volume 99, pages 518–529, 1999

    1999

  8. [8]

    Memserve: Context caching for disaggregated llm serving with elastic memory pool.arXiv preprint arXiv:2406.17565, 2024

    Cunchen Hu, Heyang Huang, Junhao Hu, Jiang Xu, Xusheng Chen, Tao Xie, Chenxi Wang, Sa Wang, Yungang Bao, Ninghui Sun, et al. Memserve: Context caching for disaggregated llm serving with elastic memory pool.arXiv preprint arXiv:2406.17565, 2024

    arXiv 2024

Show all 47 references
  1. [9]

    Cxl-anns:software-hardware collaborative memory disaggregation and computation for billion- scale approximate nearest neighbor search

    Junhyeok Jang, Hanjin Choi, Hanyeoreum Bae, Seungjun Lee, Miryeong Kwon, and Myoungsoo Jung. Cxl-anns:software-hardware collaborative memory disaggregation and computation for billion- scale approximate nearest neighbor search. In2023 USENIX Annual Technical Conference (USENIX...

    2023

  2. [10]

    Ragcache: Efficient knowledge caching for retrieval- augmented generation.arXiv preprint arXiv:2404.12457, 2024

    Chao Jin, Zili Zhang, Xuanlin Jiang, Fangyue Liu, Xin Liu, Xuanzhe Liu, and Xin Jin. Ragcache: Efficient knowledge caching for retrieval- augmented generation.arXiv preprint arXiv:2404.12457, 2024

    2024 arXiv

  3. [11]

    Design guide- lines for high performance rdma systems

    Anuj Kalia, Michael Kaminsky, and David G Andersen. Design guide- lines for high performance rdma systems. In2016 USENIX annual technical conference (USENIX ATC 16), pages 437–450, 2016

    2016

  4. [12]

    Im- proving approximate nearest neighbor search through learned adaptive early termination

    Conglong Li, Minjia Zhang, David G Andersen, and Yuxiong He. Im- proving approximate nearest neighbor search through learned adaptive early termination. InProceedings of the 2020 ACM SIGMOD Interna- tional Conference on Management of Data, pages 2539–2554, 2020

    2020

  5. [13]

    Rolex: A scalable rdma-oriented learned key-value store for disaggre- gated memory systems

    Pengfei Li, Yu Hua, Pengfei Zuo, Zhangyu Chen, and Jiajie Sheng. Rolex: A scalable rdma-oriented learned key-value store for disaggre- gated memory systems. In21st USENIX Conference on File and Storage Technologies (FAST 23), pages 99–114, 2023

    2023

  6. [14]

    https://github.com/facebookresearch/faiss

    A library for efficient similarity search and clustering of dense vectors. https://github.com/facebookresearch/faiss

  7. [15]

    Retrievalattention: Accelerating long-context llm inference via vector retrieval.arXiv preprint arXiv:2409.10516, 2024

    Di Liu, Meng Chen, Baotong Lu, Huiqiang Jiang, Zhenhua Han, Qianxi Zhang, Qi Chen, Chengruidong Zhang, Bailu Ding, Kai Zhang, et al. Retrievalattention: Accelerating long-context llm inference via vector retrieval.arXiv preprint arXiv:2409.10516, 2024

    2024 arXiv

  8. [16]

    Outback: Fast and communication-efcient index for key-value store on disaggregated memory.PVLDB, 18(2):335 – 348, 2024

    Yi Liu, Minghao Xie, Shouqian Shi, Yuanchao Xu, Heiner Litz, and Chen Qian. Outback: Fast and communication-efcient index for key-value store on disaggregated memory.PVLDB, 18(2):335 – 348, 2024

    2024

  9. [17]

    Dex: Scalable range indexing on disaggregated memory

    Baotong Lu, Kaisong Huang, Chieh-Jan Mike Liang, Tianzheng Wang, and Eric Lo. Dex: Scalable range indexing on disaggregated memory. Proceedings of the VLDB Endowment, 17(10):2603–2616, 2024

    2024

  10. [18]

    Chime: A cache-efficient and high-performance hybrid index on disaggregated memory

    Xuchuan Luo, Jiacheng Shen, Pengfei Zuo, Xin Wang, Michael R Lyu, and Yangfan Zhou. Chime: A cache-efficient and high-performance hybrid index on disaggregated memory. InProceedings of the ACM SIGOPS 30th Symposium on Operating Systems Principles, pages 110– 126, 2024

    2024

  11. [19]

    Smart: A high-performance adap- tive radix tree for disaggregated memory

    Xuchuan Luo, Pengfei Zuo, Jiacheng Shen, Jiazhen Gu, Xin Wang, Michael R Lyu, and Yangfan Zhou. Smart: A high-performance adap- tive radix tree for disaggregated memory. In17th USENIX Symposium on Operating Systems Design and Implementation (OSDI 23), pages 553–571, 2023

    2023

  12. [20]

    Yu A Malkov and Dmitry A Yashunin. Efficient and robust approxi- mate nearest neighbor search using hierarchical navigable small world graphs.IEEE transactions on pattern analysis and machine intelligence, 42(4):824–836, 2018

    2018

  13. [21]

    Revisit- ing network support for rdma

    Radhika Mittal, Alexander Shpiner, Aurojit Panda, Eitan Zahavi, Arvind Krishnamurthy, Sylvia Ratnasamy, and Scott Shenker. Revisit- ing network support for rdma. InProceedings of the 2018 Conference of the ACM Special Interest Group on Data Communication, pages 313–326, 2018

    2018

  14. [22]

    Survey of vector database management systems.The VLDB Journal, 33(5):1591–1615, 2024

    James Jie Pan, Jianguo Wang, and Guoliang Li. Survey of vector database management systems.The VLDB Journal, 33(5):1591–1615, 2024

    2024

  15. [23]

    Mooncake: A kvcache-centric disag- gregated architecture for llm serving.arXiv preprint arXiv:2407.00079, 2024

    Ruoyu Qin, Zheming Li, Weiran He, Mingxing Zhang, Yongwei Wu, Weimin Zheng, and Xinran Xu. Mooncake: A kvcache-centric disag- gregated architecture for llm serving.arXiv preprint arXiv:2407.00079, 2024

    2024 arXiv

  16. [24]

    Revisiting kd-tree for nearest neigh- bor search

    Parikshit Ram and Kaushik Sinha. Revisiting kd-tree for nearest neigh- bor search. InProceedings of the 25th acm sigkdd international confer- ence on knowledge discovery & data mining, pages 1378–1388, 2019

    2019

  17. [25]

    Hm-ann: Efficient billion-point nearest neighbor search on heterogeneous memory

    Jie Ren, Minjia Zhang, and Dong Li. Hm-ann: Efficient billion-point nearest neighbor search on heterogeneous memory. In H. Larochelle, M. Ranzato, R. Hadsell, M.F. Balcan, and H. Lin, editors,Advances in Neural Information Processing Systems, volume 33, pages 10672–10684. Curr...

    2020

  18. [26]

    Legoos: A disseminated, distributed os for hardware resource disaggregation

    Yizhou Shan, Yutong Huang, Yilun Chen, and Yiying Zhang. Legoos: A disseminated, distributed os for hardware resource disaggregation. In13th USENIX Symposium on Operating Systems Design and Imple- mentation (OSDI 18), pages 69–87, 2018

    2018

  19. [27]

    Ditto: An elastic and adaptive memory-disaggregated caching system

    Jiacheng Shen, Pengfei Zuo, Xuchuan Luo, Yuxin Su, Jiazhen Gu, Hao Feng, Yangfan Zhou, and Michael R Lyu. Ditto: An elastic and adaptive memory-disaggregated caching system. InProceedings of the 29th Symposium on Operating Systems Principles, pages 675–691, 2023

    2023

  20. [28]

    Fusee: A fully memory- disaggregatedkey-value store

    Jiacheng Shen, Pengfei Zuo, Xuchuan Luo, Tianyi Yang, Yuxin Su, Yangfan Zhou, and Michael R Lyu. Fusee: A fully memory- disaggregatedkey-value store. In21st USENIX Conference on File and Storage Technologies (FAST 23), pages 81–98, 2023

    2023

  21. [29]

    Large language models are learnable planners for long-term recommendation

    Wentao Shi, Xiangnan He, Yang Zhang, Chongming Gao, Xinyue Li, Jizhi Zhang, Qifan Wang, and Fuli Feng. Large language models are learnable planners for long-term recommendation. InProceedings of the 47th International ACM SIGIR Conference on Research and Development in Informa...

    1903

  22. [30]

    Vexless: A serverless vector data management system using cloud functions

    Yongye Su, Yinqi Sun, Minjia Zhang, and Jianguo Wang. Vexless: A serverless vector data management system using cloud functions. Proceedings of the ACM on Management of Data, 2(3):1–26, 2024. 6

    2024

  23. [31]

    Disaggregating persis- tent memory and controlling them remotely: An exploration of passive disaggregated key-value stores

    Shin-Yeh Tsai, Yizhou Shan, and Yiying Zhang. Disaggregating persis- tent memory and controlling them remotely: An exploration of passive disaggregated key-value stores. In2020 USENIX Annual Technical Con- ference (USENIX ATC 20), pages 33–48, 2020

    2020

  24. [32]

    Semeru: A memory-disaggregated managed runtime

    Chenxi Wang, Haoran Ma, Shi Liu, Yuanqi Li, Zhenyuan Ruan, Khanh Nguyen, Michael D Bond, Ravi Netravali, Miryung Kim, and Guo- qing Harry Xu. Semeru: A memory-disaggregated managed runtime. In14th USENIX Symposium on Operating Systems Design and Imple- mentation (OSDI 20), pag...

    2020

  25. [33]

    Vector databases: What’s really new and what’s next?(vldb 2024 panel).Proceedings of the VLDB Endowment, 17(12):4505–4506, 2024

    Jianguo Wang, Eric Hanson, Guoliang Li, Yannis Papakonstantinou, Harsha Simhadri, and Charles Xie. Vector databases: What’s really new and what’s next?(vldb 2024 panel).Proceedings of the VLDB Endowment, 17(12):4505–4506, 2024

    2024

  26. [34]

    Milvus: A purpose-built vector data management system

    Jianguo Wang, Xiaomeng Yi, Rentong Guo, Hai Jin, Peng Xu, Shengjun Li, Xiangyu Wang, Xiangzhou Guo, Chengming Li, Xiaohai Xu, et al. Milvus: A purpose-built vector data management system. InProceed- ings of the 2021 International Conference on Management of Data, pages 2614–2627, 2021

    2021

  27. [35]

    Disaggregated database systems

    Jianguo Wang and Qizhen Zhang. Disaggregated database systems. InCompanion of the 2023 International Conference on Management of Data, pages 37–44, 2023

    2023

  28. [36]

    Sherman: A write-optimized distributed b+ tree index on disaggregated memory

    Qing Wang, Youyou Lu, and Jiwu Shu. Sherman: A write-optimized distributed b+ tree index on disaggregated memory. InProceedings of the 2022 international conference on management of data, pages 1033–1048, 2022

    2022

  29. [37]

    The case for distributed shared-memory databases with rdma-enabled memory disaggregation.arXiv preprint arXiv:2207.03027, 2022

    Ruihong Wang, Jianguo Wang, Stratos Idreos, M Tamer Özsu, and Walid G Aref. The case for distributed shared-memory databases with rdma-enabled memory disaggregation.arXiv preprint arXiv:2207.03027, 2022

    2022 arXiv

  30. [38]

    Rcmp: Reconstructing rdma-based memory disaggregation via cxl.ACM Transactions on Architecture and Code Optimization, 21(1):1–26, 2024

    Zhonghua Wang, Yixing Guo, Kai Lu, Jiguang Wan, Daohui Wang, Ting Yao, and Huatao Wu. Rcmp: Reconstructing rdma-based memory disaggregation via cxl.ACM Transactions on Architecture and Code Optimization, 21(1):1–26, 2024

    2024

  31. [39]

    Characterizing off-path smartnic for accelerating distributed systems

    Xingda Wei, Rongxin Cheng, Yuhan Yang, Rong Chen, and Haibo Chen. Characterizing off-path smartnic for accelerating distributed systems. In17th USENIX Symposium on Operating Systems Design and Implementation (OSDI 23), pages 987–1004, 2023

    2023

  32. [40]

    Lighttraffic: On optimizing cpu-gpu data traffic for efficient large- scale random walks

    Yipeng Xing, Yongkun Li, Zhiqiang Wang, Yinlong Xu, and John CS Lui. Lighttraffic: On optimizing cpu-gpu data traffic for efficient large- scale random walks. In2023 IEEE 39th International Conference on Data Engineering (ICDE), pages 882–895. IEEE, 2023

    2023

  33. [41]

    Motor: Enabling multi- versioning for distributed transactions on disaggregated memory

    Ming Zhang, Yu Hua, and Zhijun Yang. Motor: Enabling multi- versioning for distributed transactions on disaggregated memory. In 18th USENIX Symposium on Operating Systems Design and Implemen- tation (OSDI 24), pages 801–819, 2024

    2024

  34. [42]

    Ford: Fast one- sided rdma-based distributed transactions for disaggregated persistent memory

    Ming Zhang, Yu Hua, Pengfei Zuo, and Lurong Liu. Ford: Fast one- sided rdma-based distributed transactions for disaggregated persistent memory. In20th USENIX Conference on File and Storage Technologies (FAST 22), pages 51–68, 2022

    2022

  35. [43]

    Grip: Multi-store capacity-optimized high-performance nearest neighbor search for vector search engine

    Minjia Zhang and Yuxiong He. Grip: Multi-store capacity-optimized high-performance nearest neighbor search for vector search engine. InProceedings of the 28th ACM International Conference on Information and Knowledge Management, pages 1673–1682, 2019

    2019

  36. [44]

    Fast, approximate vector queries on very large unstructured datasets

    Zili Zhang, Chao Jin, Linpeng Tang, Xuanzhe Liu, and Xin Jin. Fast, approximate vector queries on very large unstructured datasets. In20th USENIX Symposium on Networked Systems Design and Implementation (NSDI 23), pages 995–1011, 2023

    2023

  37. [45]

    Fast vector query processing for large datasets beyond gpu memory with reordered pipelining

    Zili Zhang, Fangyue Liu, Gang Huang, Xuanzhe Liu, and Xin Jin. Fast vector query processing for large datasets beyond gpu memory with reordered pipelining. In21st USENIX Symposium on Networked Systems Design and Implementation (NSDI 24), pages 23–40, 2024

    2024

  38. [46]

    Hidpu: A dpu-oriented hybrid indexing scheme for disaggregated storage systems

    Wenbin Zhu, Zhaoyan Shen, Qian Wei, Renhai Chen, Xin Yao, Dongx- iao Yu, and Zili Shao. Hidpu: A dpu-oriented hybrid indexing scheme for disaggregated storage systems. In23rd USENIX Conference on File and Storage Technologies (FAST 25), pages 271–285, 2025

    2025

  39. [47]

    Race: one- sided rdma-conscious extendible hashing.ACM Transactions on Storage (TOS), 18(2):1–29, 2022

    Pengfei Zuo, Qihui Zhou, Jiazhao Sun, Liu Yang, Shuangwu Zhang, Yu Hua, James Cheng, Rongfeng He, and Huabing Yan. Race: one- sided rdma-conscious extendible hashing.ACM Transactions on Storage (TOS), 18(2):1–29, 2022. 7

    2022

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.