Search for: All records

The growing adoption of hardware accelerators driven by their intelligent compiler and runtime system counterparts has democratized ML services and precipitously reduced their execution times. This motivates us to shift our attention to efficiently serve these ML services under distributed settings and characterize the overheads imposed by the RPC mechanism ('RPC tax') when serving them on accelerators. The RPC implementations designed over the years implicitly assume the host CPU services the requests, and we focus on expanding such works towards accelerator-based services. While recent proposals calling for SmartNICs to take on this task are reasonable for simple kernels, serving complex ML models requires a more nuanced view to optimize both the data-path and the control/orchestration of these accelerators. We program today's commodity network interface cards (NICs) to split the control and data paths for effective transfer of control while efficiently transferring the payload to the accelerator. As opposed to unified approaches that bundle these paths together, limiting the flexibility in each of these paths, we design and implement SplitRPC - a control + data path optimizing RPC mechanism for ML inference serving. SplitRPC allows us to optimize the datapath to the accelerator while simultaneously allowing the CPU to maintain full orchestration capabilities. We implement SplitRPC on both commodity NICs and SmartNICs and demonstrate how GPU-based ML services running different compiler/runtime systems can benefit. For a variety of ML models served using different inference runtimes, we demonstrate that SplitRPC is effective in minimizing the RPC tax while providing significant gains in throughput and latency over existing kernel by-pass approaches, without requiring expensive SmartNIC devices. 

The growing adoption of hardware accelerators driven by their intelligent compiler and runtime system counterparts has democratized ML services and precipitously reduced their execution times. This motivates us to shift our attention to efficiently serve these ML services under distributed settings and characterize the overheads imposed by the RPC mechanism (‘RPC tax’) when serving them on accelerators. The RPC implementations designed over the years implicitly assume the host CPU services the requests, and we focus on expanding such works towards accelerator-based services. While recent proposals calling for SmartNICs to take on this task are reasonable for simple kernels, serving complex ML models requires a more nuanced view to optimize both the data-path and the control/orchestration of these accelerators. We program today’s commodity network interface cards (NICs) to split the control and data paths for effective transfer of control while efficiently transferring the payload to the accelerator. As opposed to unified approaches that bundle these paths together, limiting the flexibility in each of these paths, we design and implement SplitRPC - a {control + data} path optimizing RPC mechanism for ML inference serving. SplitRPC allows us to optimize the datapath to the accelerator while simultaneously allowing the CPU to maintain full orchestration capabilities. We implement SplitRPC on both commodity NICs and SmartNICs and demonstrate how GPU-based ML services running different compiler/runtime systems can benefit. For a variety of ML models served using different inference runtimes, we demonstrate that SplitRPC is effective in minimizing the RPC tax while providing significant gains in throughput and latency over existing kernel by-pass approaches, without requiring expensive SmartNIC devices. 

The infinite capacity of cloud computing is an illusion: in reality, cloud providers cannot always have enough capacity of the right type, in the right place, at the right time to meet all demand. Consequently, cloud providers need to implement admission-control policies to ensure accepted capacity requests experience high availability. However, admission control in the public cloud is hard due to dynamic changes in both supply and demand: hardware might become unavailable, and actual VM consumption could vary for a variety of reasons including tenant scale-outs and fulfillment of VM reservations made by customers ahead of time. In this paper, we design and implement Kerveros, a flexible admission-control system that has three desired properties: i) high computational scalability to handle a large inventory, ii) accurate capacity provisioning for high VM availability, and iii) good packing efficiency to optimize resource usage. To achieve this, Kerveros uses novel bookkeeping techniques to quickly estimate the capacity available for incoming VM requests. Our system has been deployed in Microsoft Azure. Results from both simulations and production confirm that Kerveros achieves more than four nines of availability while sustaining request processing latencies of a few milliseconds. 

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.