With the technology trend of hardware and workload consolidation for embedded systems and the rapid development of edge computing, there has been increasing interest in supporting parallel real-time tasks to better utilize the multi-core platforms while meeting the stringent real-time constraints. For parallel real-time tasks, the federated scheduling paradigm, which assigns each parallel task a set of dedicated cores, achieves good theoretical bounds by ensuring exclusive use of processing resources to reduce interferences. However, because cores share the last-level cache and memory bandwidth resources, in practice tasks may still interfere with each other despite executing on dedicated cores. Such resource interferences due to concurrent accesses can be even more severe for embedded platforms or edge servers, where the computing power and cache/memory space are limited. To tackle this issue, in this work, we present a holistic resource allocation framework for parallel real-time tasks under federated scheduling. Under our proposed framework, in addition to dedicated cores, each parallel task is also assigned with dedicated cache and memory bandwidth resources. Further, we propose a holistic resource allocation algorithm that well balances the allocation between different resources to achieve good schedulability. Additionally, we provide a full implementation of our framework by extending the federated scheduling system with Intel’s Cache Allocation Technology and MemGuard. Finally, we demonstrate the practicality of our proposed framework via extensive numerical evaluations and empirical experiments using real benchmark programs.
more »
« less
Libra and the Art of Task Sizing in Big-Data Analytic Systems
Despite extensive investigation of job scheduling in data-intensive
computation frameworks, less consideration has been given to
optimizing job partitioning for resource utilization and efficient
processing. Instead, partitioning and job sizing are a form of dark
art, typically left to developer intuition and trial-and-error style
experimentation.
In this work, we propose that just as job scheduling and resource
allocation are out-sourced to a trusted mechanism external to the
workload, so too should be the responsibility for partitioning data
as a determinant for task size. Job partitioning essentially involves
determining the partition sizes to match the resource allocation at
the finest granularity. This is a complex, multi-dimensional problem
that is highly application specific: resource allocation, computational
runtime, shuffle and reduce communication requirements,
and task startup overheads all have strong influence on the most effective
task size for efficient processing. Depending on the partition
size, the job completion time can differ by as much as 10 times!
Fortunately, we observe a general trend underlying the tradeoff
between full resource utilization and system overhead across different
settings. The optimal job partition size balances these two
conflicting forces. Given this trend, we design Libra to automate
job partitioning as a framework extension. We integrate Libra with
Spark and evaluate its performance on EC2. Compared to state-of-the-art techniques, Libra can reduce the individual job execution
time by 25% to 70%.
more »
« less
- Award ID(s):
- 1815115
- NSF-PAR ID:
- 10122203
- Date Published:
- Journal Name:
- Symposium on Cloud Computing
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
null (Ed.)Deep neural networks (DNNs) are increasingly used for real-time inference, requiring low latency, but require significant computational power as they continue to increase in complexity. Edge clouds promise to offer lower latency due to their proximity to end-users and having powerful accelerators like GPUs to provide the computation power needed for DNNs. But it is also important to ensure that the edge-cloud resources are utilized well. For this, multiplexing several DNN models through spatial sharing of the GPU can substantially improve edge-cloud resource usage. Typical GPU runtime environments have significant interactions with the CPU, to transfer data to the GPU, for CPU-GPU synchronization on inference task completions, etc. These result in overheads. We present a DNN inference framework with a set of software primitives that reduce the overhead for DNN inference, increase GPU utilization and improve performance, with lower latency and higher throughput. Our first primitive uses the GPU DMA effectively, reducing the CPU cycles spent to transfer the data to the GPU. A second primitive uses asynchronous ‘events’ for faster task completion notification. GPU runtimes typically preclude fine-grained user control on GPU resources, causing long GPU downtimes when adjusting resources. Our third primitive supports overlapping of model-loading and execution, thus allowing GPU resource re-allocation with very little GPU idle time. Our other primitives increase inference throughput by improving scheduling and processing more requests. Overall, our primitives decrease inference latency by more than 35% and increase DNN throughput by 2-3×.more » « less
-
Apache Mesos, a cluster-wide resource manager, is widely deployed in massive scale at several Clouds and Data Centers. Mesos aims to provide high cluster utilization via fine grained resource co-scheduling and resource fairness among multiple users through Dominant Resource Fairness (DRF) based allocation. DRF takes into account different resource types (CPU, Memory, Disk I/O) requested by each application and determines the share of each cluster resource that could be allocated to the applications. Mesos has adopted a two-level scheduling policy: (1) DRF to allocate resources to competing frameworks and (2) task level scheduling by each framework for the resources allocated during the previous step. We have conducted experiments in a local Mesos cluster when used with frameworks such as Apache Aurora, Marathon, and our own framework Scylla, to study resource fairness and cluster utilization. Experimental results show how informed decision regarding second level scheduling policy of frameworks and attributes like offer holding period, offer refusal cycle and task arrival rate can reduce unfair resource distribution. Bin-Packing scheduling policy on Scylla with Marathon can reduce unfair allocation from 38% to 3%. By reducing unused free resources in offers we bring down the unfairness from to 90% to 28%. We also show the effect of task arrival rate to reduce the unfairness from 23% to 7%.more » « less
-
Obeid, Iyad ; Selesnick, Ivan ; Picone, Joseph (Ed.)The goal of this work was to design a low-cost computing facility that can support the development of an open source digital pathology corpus containing 1M images [1]. A single image from a clinical-grade digital pathology scanner can range in size from hundreds of megabytes to five gigabytes. A 1M image database requires over a petabyte (PB) of disk space. To do meaningful work in this problem space requires a significant allocation of computing resources. The improvements and expansions to our HPC (highperformance computing) cluster, known as Neuronix [2], required to support working with digital pathology fall into two broad categories: computation and storage. To handle the increased computational burden and increase job throughput, we are using Slurm [3] as our scheduler and resource manager. For storage, we have designed and implemented a multi-layer filesystem architecture to distribute a filesystem across multiple machines. These enhancements, which are entirely based on open source software, have extended the capabilities of our cluster and increased its cost-effectiveness. Slurm has numerous features that allow it to generalize to a number of different scenarios. Among the most notable is its support for GPU (graphics processing unit) scheduling. GPUs can offer a tremendous performance increase in machine learning applications [4] and Slurm’s built-in mechanisms for handling them was a key factor in making this choice. Slurm has a general resource (GRES) mechanism that can be used to configure and enable support for resources beyond the ones provided by the traditional HPC scheduler (e.g. memory, wall-clock time), and GPUs are among the GRES types that can be supported by Slurm [5]. In addition to being able to track resources, Slurm does strict enforcement of resource allocation. This becomes very important as the computational demands of the jobs increase, so that they have all the resources they need, and that they don’t take resources from other jobs. It is a common practice among GPU-enabled frameworks to query the CUDA runtime library/drivers and iterate over the list of GPUs, attempting to establish a context on all of them. Slurm is able to affect the hardware discovery process of these jobs, which enables a number of these jobs to run alongside each other, even if the GPUs are in exclusive-process mode. To store large quantities of digital pathology slides, we developed a robust, extensible distributed storage solution. We utilized a number of open source tools to create a single filesystem, which can be mounted by any machine on the network. At the lowest layer of abstraction are the hard drives, which were split into 4 60-disk chassis, using 8TB drives. To support these disks, we have two server units, each equipped with Intel Xeon CPUs and 128GB of RAM. At the filesystem level, we have implemented a multi-layer solution that: (1) connects the disks together into a single filesystem/mountpoint using the ZFS (Zettabyte File System) [6], and (2) connects filesystems on multiple machines together to form a single mountpoint using Gluster [7]. ZFS, initially developed by Sun Microsystems, provides disk-level awareness and a filesystem which takes advantage of that awareness to provide fault tolerance. At the filesystem level, ZFS protects against data corruption and the infamous RAID write-hole bug by implementing a journaling scheme (the ZFS intent log, or ZIL) and copy-on-write functionality. Each machine (1 controller + 2 disk chassis) has its own separate ZFS filesystem. Gluster, essentially a meta-filesystem, takes each of these, and provides the means to connect them together over the network and using distributed (similar to RAID 0 but without striping individual files), and mirrored (similar to RAID 1) configurations [8]. By implementing these improvements, it has been possible to expand the storage and computational power of the Neuronix cluster arbitrarily to support the most computationally-intensive endeavors by scaling horizontally. We have greatly improved the scalability of the cluster while maintaining its excellent price/performance ratio [1].more » « less
-
The property of (quasi-)reversibility of Markov chains have led to elegant characterization of steady-state distribution for complex queueing networks, e.g. celebrated Jackson networks, BCMP (Baskett, Chandi, Muntz, Palacois) and Kelly theorem. In a nutshell, despite the complicated interaction, in the steady-state, the queues in such networks exhibit independence and subsequently lead to explicit calculations of distributional properties of the queuing network that may seem impossible at the outset. The model of stochastic processing network (cf. Harrison 2000) captures variety of dynamic resource allocation problems including the flow-level networks used for modeling bandwidth sharing in the Internet, switched networks (cf. Shah, Wischik 2006) for modeling packet scheduling in the Internet router and wireless medium access, and hybrid flow-packet networks for modeling job-and-packet level scheduling in data centers. Unlike before, an appropriate resource allocation or scheduling policy is required in such networks to achieve good performance. Given the complexity, asymptotic analytic approaches such as fluid model or Lyapunov-Foster criteria to establish positive-recurrence and heavy traffic or diffusion approximation to characterize the scaled steady-state distribution became method of choice. A remarkable progress has been made along these lines over the past few decades, but there is a need for much more to match the explicit calculations in the context of reversible networks. In this work, we will present an alternative to this approach that leads to non-asymptotic, explicit characterization of steady-state distribution akin BCMP / Kelly theorems. This involves (a) identifying a "relaxation" of the given stochastic processing network in terms of an appropriate (quasi-)reversible queueing network, and (b) finding a resource allocation or scheduling policy of interest that "emulates" the "relaxed" networks within "small error". The proof is in the puddling -- we will present three examples of this program: (i) distributed scheduling in wireless network, (ii) scheduling in switched networks, and (iii) flow-packet scheduling in a data center. The notion of "baseline performance" (cf. Harrison, Mandayam, Shah, Yang 2014) will naturally emerges as a consequence of this program. We will discuss open questions pertaining multi-hop networks and computation complexity.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- https://doi.org/10.1145/3357223.3362720
