Experience shows that on today's high performance systems the utilization of different acceleration cards in conjunction with a high utilization of all other parts of the system is difficult. Future architectures, like exascale clusters, are expected to aggravate this issue as the number of cores are expected to increase and memory hierarchies are expected to become deeper. One big aspect for distributed applications is to guarantee high utilization of all available resources, including local or remote acceleration cards on a cluster while fully using all the available CPU resources and the integration of the GPU work into the overall programming model. For the integration of CUDA code we extended HPX, a general purpose C++ run time system for parallel and distributed applications of any scale, and enabled asynchronous data transfers from and to the GPU device and the asynchronous invocation of CUDA kernels on this data. Both operations are well integrated into the general programming model of HPX which allows to seamlessly overlap any GPU operation with work on the main cores. Any user defined CUDA kernel can be launched on any (local or remote) GPU device available to the distributed application. We present asynchronous implementations for the data transfers and kernel launches for CUDA code as part of a HPX asynchronous execution graph. Using this approach we can combine all remotely and locally available acceleration cards on a cluster to utilize its full performance capabilities. Overhead measurements show, that the integration of the asynchronous operations (data transfer + launches of the kernels) as part of the HPX execution graph imposes no additional computational overhead and significantly eases orchestrating coordinated and concurrent work on the main cores and the used GPU devices.
more »
« less
Efficient, Consistent Distributed Computation with Predictive Treaties
To achieve good performance, modern applications often partition and replicate their state across multiple geographically-distributed nodes. While this approach reduces latency in the common case, it can be challenging for programmers to use correctly, especially in applications that require strong consistency. We show how to achieve strong consistency while avoiding coordination by using predictive treaties, a mechanism that can significantly reduce distributed coordination without losing strong consistency. The central insight behind our approach is that many computations can be expressed in terms of predicates over distributed state that can be partitioned and enforced locally. Predictive treaties improve on previous work by allowing the locally enforced predicates to depend on time. Intuitively, by predicting the evolution of system state, coordination can be significantly reduced compared to static approaches. We implemented predictive treaties in a distributed system that exposes them via an intuitive programming model. We evaluate performance on several benchmarks, including TPC-C, showing that predictive treaties can significantly increase performance by orders of magnitude and can even outperform customized algorithms.
more »
« less
- Award ID(s):
- 1717554
- NSF-PAR ID:
- 10095544
- Date Published:
- Journal Name:
- EuroSys 2019
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
null (Ed.)We introduce consistency-aware durability or Cad, a new approach to durability in distributed storage that enables strong consistency while delivering high performance. We demonstrate the efficacy of this approach by designing cross-client monotonic reads, a novel and strong consistency property that provides monotonic reads across failures and sessions in leader-based systems; such a property can be particularly beneficial in geo-distributed and edge-computing scenarios. We build Orca, a modified version of ZooKeeper that implements Cad and cross-client monotonic reads. We experimentally show that Orca provides strong consistency while closely matching the performance of weakly consistent ZooKeeper. Compared to strongly consistent ZooKeeper, Orca provides significantly higher throughput (1.8--3.3×) and notably reduces latency, sometimes by an order of magnitude in geo-distributed settings. We also implement Cad in Redis and show that the performance benefits are similar to that of Cad’s implementation in ZooKeeper.more » « less
-
High-level data types are often associated with semantic invariants that must be preserved by any correct implementation. While having implementations enforce strong guarantees such as linearizability or serializability can often be used to prevent invariant violations in concurrent settings, such mechanisms are impractical in geo-distributed replicated environments, the platform of choice for many scalable Web services. To achieve high-availability essential to this domain, these environments admit various forms of weak consistency that do not guarantee all replicas have a consistent view of an application's state. Consequently, they often admit difficult-to-understand anomalous behaviors that violate a data type's invariants, but which are extremely challenging, even for experts, to understand and debug. In this paper, we propose a novel programming framework for replicated data types (RDTs) equipped with an automatic (bounded) verification technique that discovers and fixes weak consistency anomalies. Our approach, implemented in a tool called Q9, involves systematically exploring the state space of an application executing on top of an eventually consistent data store, under an unrestricted consistency model but with a finite concurrency bound. Q9 uncovers anomalies (i.e., invariant violations) that manifest as finite counterexamples, and automatically generates repairs for such anamolies by selectively strengthening consistency guarantees for specific operations. Using Q9, we have uncovered a range of subtle anomalies in implementations of well-known benchmarks, and have been able to apply the repairs it mandates to effectively eliminate them. Notably, these benchmarks were written adopting best practices suggested to manage distributed replicated state (e.g., they are composed of provably convergent RDTs (CRDTs), avoid mutable state, etc.). While the safety guarantees offered by our technique are constrained by the concurrency bound, we show that in practice, proving bounded safety guarantees typically generalize to the unbounded case.more » « less
-
Ceph is an open source distributed storage system that is object-based and massively scalable. Ceph provides developers with the capability to create data interfaces that can take advantage of local CPU and memory on the storage nodes (Ceph Object Storage Devices). These interfaces are powerful for application developers and can be created in C, C++, and Lua. Skyhook is an open source storage and database project in the Center for Research in Open Source Software at UC Santa Cruz. Skyhook uses these capabilities in Ceph to create specialized read/write interfaces that leverage IO and CPU within the storage layer toward database processing and management. Specifically, we develop methods to apply predicates locally as well as additional metadata and indexing capabilities using Ceph's internal indexing mechanism built on top of RocksDB. Skyhook's approach helps to enable scale-out of a single node database system by scaling out the storage layer. Our results show the performance benefits for some queries indeed scale well as the storage layer scales out.more » « less
-
Millions of sensors, mobile applications and machines now generate billions of events. Specialized many-core key-value stores (KVSs) can ingest and index these events at high rates (over 100 Mops/s on one machine) if events are generated on the same machine; however, to be practical and cost-effective they must ingest events over the network and scale across cloud resources elastically. We present Shadowfax, a new distributed KVS based on FASTER, that transparently spans DRAM, SSDs, and cloud blob storage while serving 130 Mops/s/VM over commodity Azure VMs using conventional Linux TCP. Beyond high single-VM performance, Shadowfax uses a unique approach to distributed reconfiguration that avoids any server-side key ownership checks or cross-core coordination both during normal operation and migration. Hence, Shadowfax can shift load in 17 s to improve system throughput by 10 Mops/s with little disruption. Compared to the state-of-the-art, it has 8x better throughput (than Seastar+memcached) and avoids costly I/O to move cold data during migration. On 12 machines, Shadowfax retains its high throughput to perform 930 Mops/s, which, to the best of our knowledge, is the highest reported throughput for a distributed KVS used for large-scale data ingestion and indexing.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- The DOI is not currently available.
