Multi-toroidal Interconnects: Using Additional Communication Links to Improve Utilization of Parallel Computers

Aridor, Yariv; Domany, Tamar; Goldshmidt, O.; Shmueli, Edi; Moreira, José E.; Stockmeier, Larry

doi:10.1007/11407522_8

Cited by 5 publications

(13 citation statements)

References 9 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This clearly indicates the advantage of the multitoroidal topology over the 3D torus for allocation of toroidal partitions, at least for the baseline allocation scheme, confirming the conclusions in [8]. Note that we made no changes in the allocation algorithm, and the algorithm is not even optimized for the switch-based architecture.…”

Section: Tori Allocationsupporting

confidence: 87%

“…We show that multitoroidal interconnects achieve significantly better (sometimes by a factor of 2 or more) resource utilization than toroidal ones (see also [8]), allow multihop allocation of isolated partitions that helps decrease fragmentation (cf. Section 4.2), and are much less susceptible to link failures (cf.…”

Section: Multitoroidal Topologymentioning

confidence: 98%

“…Our earlier work [8], [31] discussed specific examples of multitoroidal topology and very particular and limited allocation schemes. This paper develops a robust generic methodology for comparison between traditional and multitoroidal architectures, including a comprehensive price and performance analysis of generic resource allocation algorithms.…”

Section: Related Workmentioning

confidence: 99%

“…In [8], we proposed augmenting the toroidal line in Fig. 3b with additional links connecting the allocation units within the line.…”

Section: The Topology Of a Linementioning

confidence: 99%

“…8a, any set of continuously indexed mesh or torus partitions can be coallocated (cf. [8]). However, we lose some options because nodes 0 and 7 are not adjacent.…”

Section: Multitoroidal Architecturementioning

confidence: 99%

See 4 more Smart Citations

Multitoroidal Interconnects For Tightly Coupled Supercomputers

Aridor

Domany²,

Goldshmidt³

et al. 2008

IEEE Trans. Parallel Distrib. Syst.

View full text Add to dashboard Cite

The processing elements of many modern tightly coupled multicomputers are connected via mesh or toroidal networks. Such interconnects are simple and highly scalable, but suffer from high fragmentation, low utilization, and insufficient fault tolerance when the resources allocated to each job are dedicated. High-dimensional interconnects may be more efficient in certain cases, but are based on complex and expensive components and scale poorly. We present a novel hardware/software architectural approach that detaches the processing elements of the system from the interconnect and augments the traditional toroidal topology to provide additional connectivity options and additional link redundancy. We explore the properties of the new "multitoroidal" topology and the improvements it offers in resource utilization and failure tolerance. We present the results of extensive simulation studies to show that for practically important types of workloads, the resource utilization may be increased by 50 percent and, in certain cases, as much as 100 percent compared to toroidal machines and is, in fact, close to the theoretically optimal case of a full crossbar interconnect. The combined hardware/software architectural innovation is a major significant improvement in resource utilization on top of the state of the art in scheduling algorithm research. Also, multitoroidal multicomputers are able to work under link failure rates of 0.002 failures per week that would shut down toroidal machines. A variant of the multitoroidal architecture is implemented in the Blue Gene/L supercomputer.

show abstract

Section: Tori Allocationsupporting

confidence: 87%

Section: Multitoroidal Topologymentioning

confidence: 98%

Section: Related Workmentioning

confidence: 99%

“…In [8], we proposed augmenting the toroidal line in Fig. 3b with additional links connecting the allocation units within the line.…”

Section: The Topology Of a Linementioning

confidence: 99%

“…8a, any set of continuously indexed mesh or torus partitions can be coallocated (cf. [8]). However, we lose some options because nodes 0 and 7 are not adjacent.…”

Section: Multitoroidal Architecturementioning

confidence: 99%

See 3 more Smart Citations

Multitoroidal Interconnects For Tightly Coupled Supercomputers

Aridor

Domany²,

Goldshmidt³

et al. 2008

IEEE Trans. Parallel Distrib. Syst.

View full text Add to dashboard Cite

show abstract

Open Job Management Architecture for the Blue Gene/L Supercomputer

Aridor

Domany

Goldshmidt

et al. 2005

Job Scheduling Strategies for Parallel Processing

View full text Add to dashboard Cite

Efficient subtorus processor allocation in a multi-dimensional torus

Mao

Chen

Watson

2005

Eighth International Conference on High-Performance Computing in Asia-Pacific Region (HPCASIA'05)

View full text Add to dashboard Cite

Processor allocation in a mesh or torus connected multicomputer system with up to three dimensions is a hard problem that has received some research attention in the past decade. With the recent deployment of multicomputer systems with a torus topology of dimensions higher than three, which are used to solve complex problems arising in scientific computing, it becomes imminent to study the problem of allocating processors of the configuration of a torus in a multi-dimensional torus connected system. In this paper, we first define the concept of a semitorus. We present two partition schemes, the Equal Partition (EP) and the Non-Equal Partition (NEP), that partition a multi-dimensional semitorus into a set of sub-semitori. We then propose two processor allocation algorithms based on these partition schemes. We evaluate our algorithms by incorporating them in commonly used FCFS and backfilling scheduling policies and conducting simulation using workload traces from the Parallel Workloads Archive. Specifically, our simulation experiments compare four algorithm combinations, FCFS/EP, FCFS/NEP, backfilling/EP, and backfilling/NEP, for two existing multi-dimensional torus connected systems. The simulation results show that our algorithms (especially the backfilling/NEP combination) are capable of producing schedules with system utilization and mean job bounded slowdowns comparable to those in a fully connected multicomputer.

show abstract

Multi-toroidal Interconnects: Using Additional Communication Links to Improve Utilization of Parallel Computers

Cited by 5 publications

References 9 publications

Multitoroidal Interconnects For Tightly Coupled Supercomputers

Multitoroidal Interconnects For Tightly Coupled Supercomputers

Open Job Management Architecture for the Blue Gene/L Supercomputer

Efficient subtorus processor allocation in a multi-dimensional torus

Contact Info

Product

Resources

About