Model-serving systems expose machine learning (ML) models to applications programmatically via a high-level API. Cloud plat- forms use these systems to mask the complexities of optimally managing resources and servicing inference requests across multi- ple applications. Model serving at the edge is now also becoming increasingly important to support inference workloads with tight latency requirements. However, edge model serving differs substan- tially from cloud model serving in its latency, energy, and accuracy constraints: these systems must support multiple applications with widely different latency and accuracy requirements on embedded edge accelerators with limited computational and energy resources.
To address the problem, this paper presents Dělen,1 a flexible and adaptive model-serving system for multi-tenant edge AI. Dělen exposes a high-level API that enables individual edge applications to specify a bound at runtime on the latency, accuracy, or energy of their inference requests. We efficiently implement Dělen using conditional execution in multi-exit deep neural networks (DNNs), which enables granular control over inference requests, and evalu- ate it on a resource-constrained Jetson Nano edge accelerator. We evaluate Dělen flexibility by implementing state-of-the-art adapta- tion policies using Dělen’s API, and evaluate its adaptability under different workload dynamics and goals when running single and multiple applications.
more »
« less
The Fast and The Frugal: Tail Latency Aware Provisioning for Coping with Load Variations
Small and medium sized enterprises use the cloud for running online, user-facing, tail latency sensitive applications with well-defined fixed monthly budgets. For these applications, adequate system capacity must be provisioned to extract maximal performance despite the challenges of uncertainties in load and request-sizes. In this paper, we address the problem of capacity provisioning under fixed budget constraints with the goal of minimizing tail latency.
To tackle this problem, we propose building systems using a heterogeneous mix of low latency expensive resources and cheap resources that provide high throughput per dollar. As load changes through the day, we use more faster resources to reduce tail latency during low load periods and more cheaper resources to handle the high load periods. To achieve these tail latency benefits, we introduce novel heterogeneity-aware scheduling and autoscaling algorithms that are designed for minimizing tail latency. Using software prototypes and by running experiments on the public cloud, we show that our approach can outperform existing capacity provisioning systems by reducing the tail latency by as much as 45% under fixed-budget settings.
more »
« less
- PAR ID:
- 10173263
- Date Published:
- Journal Name:
- WWW '20: Proceedings of The Web Conference 2020
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
The increased use of micro-services to build web applications has spurred the rapid growth of Function-as-a-Service (FaaS) or serverless computing platforms. While FaaS simplifies provisioning and scaling for application developers, it introduces new challenges in resource management that need to be handled by the cloud provider. Our analysis of popular serverless workloads indicates that schedulers need to handle functions that are very short-lived, have unpredictable arrival patterns, and require expensive setup of sandboxes. The challenge of running a large number of such functions in a multi-tenant cluster makes existing scheduling frameworks unsuitable. We present Archipelago, a platform that enables low latency request execution in a multi-tenant serverless setting. Archipelago views each application as a DAG of functions, and every DAG in associated with a latency deadline. Archipelago achieves its per-DAG request latency goals by: (1) partitioning a given cluster into a number of smaller worker pools, and associating each pool with a semi-global scheduler (SGS), (2) using a latency-aware scheduler within each SGS along with proactive sandbox allocation to reduce overheads, and (3) using a load balancing layer to route requests for different DAGs to the appropriate SGS, and automatically scale the number of SGSs per DAG. Our testbed results show that Archipelago meets the latency deadline for more than 99% of realistic application request workloads, and reduces tail latencies by up to 36X compared to state-of-the-art serverless platforms.more » « less
-
Ease of use and transparent access to elastic resources have attracted many applications away from traditional platforms toward serverless functions. Many of these applications, such as machine learning, could benefit significantly from GPU acceleration. Unfortunately, GPUs remain inaccessible from serverless functions in modern production settings. We present DGSF, a platform that transparently enables serverless functions to use GPUs through general purpose APIs such as CUDA. DGSF solves provisioning and utilization challenges with disaggregation, serving the needs of a potentially large number of functions through virtual GPUs backed by a small pool of physical GPUs on dedicated servers. Disaggregation allows the provider to decouple GPU provisioning from other resources, and enables significant benefits through consolidation. We describe how DGSF solves GPU disaggregation challenges including supporting API transparency, hiding the latency of communication with remote GPUs, and load-balancing access to heavily shared GPUs. Evaluation of our prototype on six workloads shows that DGSF’s API remoting optimizations can improve the runtime of a function by up to 50% relative to unoptimized DGSF. Such optimizations, which aggressively remove GPU runtime and object management latency from the critical path, can enable functions running over DGSF to have a lower end-to-end time than when running on a GPU natively. By enabling GPU sharing, DGSF can reduce function queueing latency by up to 53%. We use DGSF to augment AWS Lambda with GPU support, showing similar benefits.more » « less
-
The slowdown of Moore’s Law, combined with advances in 3D stacking of logic and memory, have pushed architects to revisit the concept of processing-in-memory (PIM) to overcome the memory wall bottleneck. This PIM renaissance finds itself in a very different computing landscape from the one twenty years ago, as more and more computation shifts to the cloud. Most PIM architecture papers still focus on best-effort applications, while PIM’s impact on latency-critical cloud applications is not well understood. This paper explores how datacenters can exploit PIM architectures in the context of latency-critical applications. We adopt a general-purpose cloud server with HBM-based, 3D-stacked logic+memory modules, and study the impact of PIM on six diverse interactive cloud applications. We reveal the previously neglected opportunity that PIM presents to these services, and show the importance of properly managing PIM-related resources to meet the QoS targets of interactive services and maximize resource efficiency. Then, we present PIMCloud, a QoS-aware resource manager designed for cloud systems with PIM allowing colocation of multiple latency-critical and best-effort applications. We show that PIMCloud efficiently manages PIM resources: it (1) improves effective machine utilization by up to 70% and 85% (average 24% and 33%) under 2-app and 3-app mixes, compared to the best state-of-the-art manager; (2) helps latency-critical applications meet QoS; and (3) adapts to varying load patterns.more » « less
-
Maintaining low tail latency is critical for the efficiency and performance of large-scale datacenter systems. Software bugs that cause tail latency problems, however, are notoriously difficult to debug. We present LDB, a new latency profiling tool that aims to overcome this challenge by precisely identifying the specific functions that are responsible for tail latency anomalies. LDB observes the latency of all functions in a running program. It uses a novel, software-only technique called stack sampling, where a busy-spinning stack scanner thread polls lightweight metadata recorded in the call stack, shifting tracing costs away from program threads. In addition, LDB uses event tagging to record requests, inter-thread synchronization, and context switching. This can be used, for example, to generate per-request timelines and to find the root cause of complex tail latency problems such as lock contention in multi-threaded programs. We evaluate LDB with three datacenter applications, finding latency problems in each. Our results further show that LDB produces actionable insights, has low overhead, and can rapidly analyze recordings, making it feasible to use in production settings.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- https://doi.org/10.1145/3366423
