Machine learning (ML) frameworks, such as PyTorch and TensorFlow, rely on data loaders to preprocess data before feeding it to accelerators. When preprocessing is inefficiently pipelined, GPUs can remain idle over long periods of time, leading to substantial training delays. For example, PyTorch's default data loaders can cause up to 76% GPU idleness. A key bottleneck is the variability in preprocessing time across samples within the same dataset. Existing data loaders are oblivious to this variability, training all samples uniformly. In this case, a single slow sample can stall the entire batch, causing head-of-line blocking.

Modern data-intensive systems rely on in-memory caching to achieve high throughput and low latency. CacheLib, Meta's general-purpose caching engine, provides high performance and flexibility for building specialized caches for a variety of applications. However, despite its wide adoption in large-scale infrastructures, CacheLib's data management mechanisms exhibit inefficiencies in shared environments. Particularly, its static and uncoordinated memory allocation leads to fragmented resource usage, unfair memory distribution, and degraded performance across tenants and instances.

We present Keigo, a concurrency- and workload-aware storage middleware that enhances the performance of log-structured merge key-value stores (LSM KVS) when they are deployed on a hierarchy of storage devices. The key observation behind Keigo is that there is no one-size-fits-all placement of data across the storage hierarchy that optimizes for all workloads. Hence, to leverage the benefits of combining different storage devices, Keigo places files across different devices based on their parallelism, I/O bandwidth, and capacity. We introduce three techniques - concurrency-aware data placement, persistent read-only caching , and context-based I/O differentiation. Keigo is portable across different LSMs, is adaptable to dynamic workloads, and does not require extensive profiling. Our system enables established production KVS such as RocksDB, LevelDB, and Speedb to benefit from heterogeneous storage setups. We evaluate Keigo using synthetic and realistic workloads, showing that it improves the throughput of production-grade LSMs up to 4X for write- and 18X for read-heavy workloads when compared to general-purpose storage systems and specialized LSM KVS.

We present LazyFS, a new fault injection tool that simplifies the debugging and reproduction of complex data durability bugs experienced by databases, key-value stores, and other data-centric systems in crashes. Our tool simulates persistence properties of POSIX file systems (e.g., operations ordering and atomicity) and enables users to inject lost and torn write faults with a precise and controlled approach. Further, it provides profiling information about the system's operations flow and persisted data, enabling users to better understand the root cause of errors. We use LazyFS to study seven important systems: PostgreSQL, etcd, Zookeeper, Redis, LevelDB, PebblesDB, and Lightning Network. Our fault injection campaign shows that LazyFS automates and facilitates the reproduction of five known bug reports containing manual and complex reproducibility steps. Further, it aids in understanding and reproducing seven ambiguous bugs reported by users. Finally, LazyFS is used to find eight new bugs, which lead to data loss, corruption, and unavailability.

Modern I/O applications that run on HPC infrastructures are increasingly becoming read and metadata intensive. However, having multiple applications submitting large amounts of metadata operations can easily saturate the shared parallel file system's metadata resources, leading to overall performance degradation and I/O unfairness. We present PADLL, an application and file system agnostic storage middleware that enables QoS control of data and metadata workflows in HPC storage systems. It adopts ideas from Software-Defined Storage, building data plane stages that mediate and rate limit POSIX requests submitted to the shared file system, and a control plane that holistically coordinates how all I/O workflows are handled. We demonstrate its performance and feasibility under multiple QoS policies using synthetic benchmarks, real-world applications, and traces collected from a production file system. Results show that PADLL can enforce complex storage QoS policies over concurrent metadata-aggressive jobs, ensuring fairness and prioritization.

We present Monarch,a framework-agnostic storage middleware that transparently employs storage tiering to accelerate Deep Learning (DL) training. It leverages existing storage tiers of modern supercomputers (i.e., compute node's local storage and shared parallel file system (PFS)), while considering the I/O patterns of DL frameworks to improve data placement across tiers. Monarchaims at accelerating DL training and decreasing the I/O pressure imposed over the PFS. We apply Monarchto TensorFlow and PyTorch, while validating its performance and applicability under different models and dataset sizes. Results show that, even when the training dataset can only be partially stored at local storage, Monarchreduces TensorFlow's and PyTorch's training time by up to 28% and 37% for I/O-intensive models, respectively. Furthermore, Monarchdecreases the number of I/O operations submitted to the PFS by up to 56%.

Modern large-scale I/O applications that run on HPC infrastructures are increasingly becoming metadata-intensive. Unfortunately, having multiple concurrent applications submitting massive amounts of metadata operations can easily saturate the shared parallel file system's metadata resources, leading to unresponsiveness of the storage backend and overall performance degradation. To address these challenges, we present Padll, a storage middleware that enables system administrators to proactively control and ensure QoS over metadata workflows in HPC storage systems. We demonstrate its performance and feasibility by controlling the rate of both synthetic and realistic I/O workloads. Results show that Padll can dynamically control metadata-aggressive workloads, prevent I/O burstiness, and ensure I/O fairness and prioritization.

Due to convenience and usability, many deep learning (DL) jobs resort to the available shared parallel file system (PFS) for storing and accessing training data when running in HPC environments. Under such a scenario, however, where multiple I/O-intensive applications operate concurrently, the PFS can quickly get saturated with simultaneous storage requests and become a critical performance bottleneck, leading to throughput variability and performance loss.We present MONARCH, a framework-agnostic middleware for hierarchical storage management. This solution leverages the existing storage tiers present at modern supercomputers (e.g., compute node’s local storage, PFS) to improve DL training performance and alleviate the current I/O pressure of the shared PFS.We validate the applicability of our approach by developing and integrating an early prototype with the TensorFlow DL framework. Results show that MONARCH can reduce I/O operations submitted to the shared PFS by up to 45%, decreasing training time by 24% and 12%, for I/O-intensive models, namely LeNet and AlexNet.

Deep Learning (DL) training requires efficient access to large collections of data, leading DL frameworks to implement individual I/O optimizations to take full advantage of storage performance. However, these optimizations are intrinsic to each framework, limiting their applicability and portability across DL solutions, while making them inefficient for scenarios where multiple applications compete for shared storage resources.We argue that storage optimizations should be decoupled from DL frameworks and moved to a dedicated storage layer. To achieve this, we propose a new Software-Defined Storage architecture for accelerating DL training performance. The data plane implements self-contained, generally applicable I/O optimizations, while the control plane dynamically adapts them to cope with workload variations and multi-tenant environments.We validate the applicability and portability of our approach by developing and integrating an early prototype with the TensorFlow and PyTorch frameworks. Results show that our I/O optimizations significantly reduce DL training time by up to 54% and 63% for TensorFlow and PyTorch baseline configurations, while providing similar performance benefits to framework-intrinsic I/O mechanisms provided by TensorFlow.

Modern general-purpose operating systems implement major parts of their storage stacks in the kernel. Although this bolsters performance, it also complicates development and stifles innovation for today's increasingly complex storage systems. In contrast, implementing system services at the user level eases development and maintenance, and leads to improved portability, reliability, fault tolerance, and security. This has motivated the widespread use in both academia and industry of frameworks such as FUSE, which enable the implementation of file systems in user space.

Cloud infrastructures provide database services as cost-efficient and scalable solutions for storing and processing large amounts of data. To maximize performance, these services require users to trust sensitive information to the cloud provider, which raises privacy and legal concerns. This represents a major obstacle to the adoption of the cloud computing paradigm. Recent work addressed this issue by extending databases to compute over encrypted data. However, these approaches usually support a single and strict combination of cryptographic techniques invariably making them application specific. To assess and broaden the applicability of cryptographic techniques in secure cloud storage and processing, these techniques need to be thoroughly evaluated in a modular and configurable database environment. This is even more noticeable for NoSQL data stores where data privacy is still mostly overlooked. In this paper, we present a generic NoSQL framework and a set of libraries supporting data processing cryptographic techniques that can be used with existing NoSQL engines and composed to meet the privacy and performance requirements of different applications. This is achieved through a modular and extensible design that enables data processing over multiple cryptographic techniques applied on the same database. For each technique, we provide an overview of its security model, along with an extensive set of experiments. The framework is evaluated with the YCSB benchmark, where we assess the practicality and performance tradeoffs for different combinations of cryptographic techniques. The results for a set of macro experiments show that the average overhead in NoSQL operations performance is below 15%, when comparing our system with a baseline database without privacy guarantees.