PTask

Supporting x86-64 address translation for 100s of GPU lanes

“…There are multiple attempts to reduce the complexity of the GPGPU programming model through software [13,32]. While these frameworks simplify the code for straightforward applications, like the UVA implementation of the vectorcopy example presented, it is still difficult to represent complex data structures.…”

Cooperative GPGPU Scheduling for Consolidating Server Workloads

“…If a GPU kernel has been running for a long time, the Gdev scheduler assigns long slices of time to other GPU app kernels to achieve fair GPU utilization. PTask [28], where a GPGPU app is designed as a data flow graph that consists of GPU kernel modules, schedules GPU kernels when they are launched. These kernel-based schedulers suffer from the same problem as the command-based ones.…”

“…Scientific apps [5], [6] exclusively use GPUs to compute their simulations. Existing GPU resource managers, including GPU command-based schedulers [24]- [26], novel GPU kernel launchers [27], [28], and thread block schedulers [29], [30], fail to schedule GPU eaters appropriately since GPU eaters do not provide scheduling points such as kernel launches or thread block completion; thus, a hosted GPU eater may monopolize the GPU. Other techniques, such as context funneling [31], [32] and persistent threads [33], effectively schedule GPU eaters but fail to isolate GPGPU apps; thus, a hosted GPGPU app may access and modify the memory of other GPGPU apps, which is not suitable for multi-tenant cloud platforms.…”

Modeling NICs with Unicorn

“…However, the problem is different to using heterogeneous cores, whether GPUs [7,14] or others [18]. NICs mostly provide fixed hardware functions rather than programmable cores, and different NIC models, even within a vendor, offer very different features and configuration options 1 .…”

“…The dataflow model of computation also has a long history, and has recently been applied in parallel programming [1,6,14]. Dataflow representations of network processing are used in Click [8] and the x-kernel [3].…”