What you can't forget

Bae, Hanyeoreum; Kim, Jiseon; Kwon, Miryeong

doi:10.1145/3538643.3539744

Cited by 25 publications

(6 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Even though Zone Append only shows high performance when write requests are issued to a single zone, it is still beneficial for scenarios where shared storage systems need performance isolation [23,32,38]. Specifically, since a zone is typically mapped to only a subset of physical flash channels or chips [17], a shared storage system based on ZNS SSDs should allocate only a limited number of open zones to applications that do not have high write bandwidth demands. This isolates the write traffic of the applications to a subset of flash channels and limits their impact on other co-located applications.…”

Section: (A))mentioning

confidence: 99%

Identity of university Chinese heritage language learners in Hong Kong

Li¹,

李蓁²

View full text Add to dashboard Cite

The core of out-of-distribution (OOD) detection is to learn the in-distribution (ID) representation, which is distinguishable from OOD samples. Previous work applied recognition-based methods to learn the ID features, which tend to learn shortcuts instead of comprehensive representations. In this work, we find surprisingly that simply using reconstruction-based methods could boost the performance of OOD detection significantly. We deeply explore the main contributors of OOD detection and find that reconstructionbased pretext tasks have the potential to provide a generally applicable and efficacious prior, which benefits the model in learning intrinsic data distributions of the ID dataset. Specifically, we take Masked Image Modeling as a pretext task for our OOD detection framework (MOOD). Without bells and whistles, MOOD outperforms previous SOTA of one-class OOD detection by 5.7%, multi-class OOD detection by 3.0%, and near-distribution OOD detection by 2.1%. It even defeats the 10-shot-per-class outlier exposure OOD detection, although we do not include any OOD samples for our detection.

show abstract

Section: (A))mentioning

confidence: 99%

Identity of university Chinese heritage language learners in Hong Kong

Li¹,

李蓁²

View full text Add to dashboard Cite

show abstract

“…Recent ZNS studies [7], [8], [12] focus on various aspects of hardware and software co-design. Traditional SSDs have internal DRAM and over-provisioning (OP) flash capacity for garbage collection, coined as block interface tax in [7], which can be minimized by Zone-based flash management.…”

Section: B Zns Ssd and Zenfsmentioning

confidence: 99%

“…Another potential benefit of ZNS SSDs is performance isolation [8]. The authors presented two kinds of ZNS SSDs: typical large-zone ZNS SSD and small-zone ZNS SSD.…”

Section: B Zns Ssd and Zenfsmentioning

confidence: 99%

“…That would work for large-zone ZNS SSD that has GB-sized zones. However, the zone reclaiming cost is considerable in large-zone ZNS SSDs [8]. For small-zone ZNS SSD, zone utilization increases relatively quickly than large-zone ZNS SSD.…”

Section: A Motivation: Broken Promise Of Zns Ssdmentioning

confidence: 99%

“…In contrast, other ZNS SSD devices have a smaller zone size (e.g., 72 MB), and a zone has limited internal parallelism. With a small-zone ZNS SSD device, reading or writing to a zone utilizes limited internal channels, presenting low bandwidth [8]. However, it is better to control bandwidth between flush and compaction because we can flexibly choose the flush zones to avoid interference from the compaction zone.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

ZenFS+: Nurturing Performance and Isolation to ZenFS

Yoo

Choi

et al. 2023

IEEE Access

View full text Add to dashboard Cite

This paper proposes ZenFS+, a new storage backend of RocksDB for small-zone ZNS SSD. RocksDB has complicated internal operations such as flush and compaction. Flush and compaction run in separate threads, and it is well known that they closely interact around the sstables. Due to the concurrent storage I/O between flush and compaction, ZenFS presents unsatisfactory performance for small-zone ZNS SSD. We believe that ZenFS+ presents how ZNS SSD can support the performance and isolation of modern keyvalue stores. Leveraging the ZNS SSD, ZenFS+ intelligently identifies the independent zones group (IZG), which reveals the device's internal parallelism. With the IZG information, ZenFS+ effectively isolates the flush workloads of RocksDB from compaction. Besides, ZenFS+ spreads the sstables to multiple IZGs so that the storage write can leverage the hardware parallelism. ZenFS+ presents up to 4.8x higher storage throughput for write-intensive microbenchmark and stabilizing 99.9P tail latency by 1/51 from existing ZenFS. Further, ZenFS+ implements proactive garbage collection, and presents sustainable performance over longer lifespan of the device.

show abstract