All shapes contiguous submesh allocation for 2D mesh multicomputers

Abstract

Although contiguous allocation is known to suffer from high processor fragmentation, it is a viable allocation approach as it isolates jobs from one another, which can improve system security. Most recent contiguous allocation schemes proposed for two-dimensional meshes switch the orientation of allocation requests when allocation fails for the initial request orientation. This transformation was found to reduce processor fragmentation and improve system performance. In this paper, we propose to generalise this restricted transformation by permitting all request shapes. Using simulations, we show that this generalisation when combined with giving priority to allocating peripheral submeshes can result in significant system performance gains.

On Allocation Algorithms for Manycore Systems With Network on Chip

“…The All-Shapes First-Fit contiguous submesh allocation strategy (ASFF) (Ababneh et al, 2010), attempts allocation to an incoming job request by permitting all possible 2D shapes. In ASFF, given a job request for n cores, all valid request shapes are constructed, and then these shapes are considered for allocation in a specific order.…”

An Efficient All Shapes Busy List Processor Allocation Algorithm for 3D Mesh Multicomputers

“…The size of the mesh, N , is the number of processors it has, where N WDH = (Ababneh et al, 2015;Bani-Mohammad et al, 2006). The following definitions have been adopted from (Chiu & Chen, 1999;Ababneh et al, 2010;Bani-Mohammad et al, 2006).…”

“…In distributed memory multicomputers, separate contiguous processor sub-meshes are assigned to jobs for the duration of their execution (Yoo & Das, 2002;Ababneh, 2006;Ababneh, 2001;Ababneh, 2009;Ababneh et al, 2015;Ababneh et al, 2010;Bani-Mohammad et al, 2006;Bani-Mohammad et al, 2007b;Bani-Mohammad et al, 2009;Zhu, 1992). Both two-dimensional (2D) and three-dimensional (3D) meshes and tori have been utilized in recent experimental and commercial multicomputers.…”

“…Both two-dimensional (2D) and three-dimensional (3D) meshes and tori have been utilized in recent experimental and commercial multicomputers. Most contiguous processor allocation strategies that are proposed in the previous studies are for 2D meshes (Chiu & Chen, 1999;Ababneh, 2006;Ababneh, 2009;Ababneh et al, 2010;Chuang & Tzeng, 1994;Zhu, 1992). In spite of the fact that the 2D mesh has been used in a various of parallel machines, such as the Cray XE6m (Cray, 2014b), the iWARP (Peterson et al, 1991) and the Touchstone Delta system (Intel, 1991), current multicomputers, such as the K-computer (Riken & Fujitsu, 2014), the Cray XE6 (Cray, 2014a) and the IBM BlueGene/L (Horn et al, 2014) use the 3D mesh or the 3D tori (Cray, 2014a;Horn et al, 2014;Riken & Fujitsu, 2014) as an interconnection network due to their characteristics such as lower diameter and lower average communication distance (Athas & Seitz, 1988).…”

“…The vast majority of the existing processor allocation strategies use contiguous sub-mesh allocation (Yoo & Das, 2002;Sharma & Pradhan, 1996;Gabrani & Mulkar, 2005;Kim & Yoon, 1998;Chiu & Chen, 1999;Choo et al, 2000;Ababneh, 2006;Ababneh, 2001;Ababneh, 2009;Ababneh et al, 2015;Ababneh et al, 2010;Bani-Mohammad et al, 2006;Bani-Mohammad et al, 2007b;Bani-Mohammad et al, 2009;Zhu, 1992), where a distinct contiguous sub-mesh of processors is allocated to a parallel job for the duration of its execution. This can lead to high external processor fragmentation that happens when the processor allocation strategy cannot allocate available sub-mesh of processors to an incoming job because the available processors are not contiguous or they do not have the same topology as the interconnection system network.…”