Advances, Systems and Applications
Table 2 Summary of related work on container-based HPC
Paper | Year | Objective | Containers Run-time | Benchmarks Used | Type of Physical Environment | Conclusion/Remarks | |
|---|---|---|---|---|---|---|---|
Single-server | Cluster | ||||||
Migual et al. [7] | 2013 | performance evaluation and isolation of container-based virtualization | LXC, Linix-VServer, OpenVZ | LINPACK, STREAM, NAS ParallelBenchmarks (NPB), IOzone, Isolation Benchmark Suite(IBS),NetPIPE | âś“ | âś“ | container-based virtualization shows near-native performance, but isolation feature requires maturity, Presented results with limited insight |
David et al. [5] | 2015 | performance analysis of container-based solution | LXC inside virtual environment | HPL, Network Protocol Independent Performance Evaluator (NetPIPE) | âś“ | âś— | LXC presents better performance than KVM, can be treated as a viable solution for HPC, Presented results with limited insight |
Felter et al. [6] | 2015 | performance comparison between hypervisor-based virtualization and container-based virtualization | Docker | MySQL | âś“ | âś— | docker equal or exceed the performance compared to KVM but need careful optimization related to I/O in both the cases, presented results with moderate analysis |
Zhang et al. [17] | 2015 | comprehensive performance comparison of I/O virtualization between VM and container using PCI-passthrough and SR-IOV in the HPC context | Docker | IB verbs, Graph500, NAS, LAMMPS, SPEC MPI 2007 | âś— | âś“ | container-based PCI-passthrough outperform counter hypervisor-based solution, report results with moderate analysis |
Arango et al. [8] | 2017 | performance comparison of container-based HPC environment | LXC, Docker and Singularity | HPL, IOzone, STREAM, MVA-PICH OSU Micro-Benchmarks, NAMD (NAnoscale Molecular Dynamics) | âś“ | âś“ | singularity outperform among all discussed container technology but needs improvement in the network space, reported preliminary result with very little insight |
Li et al. [9] | 2017 | performance overhead analysis of container-based virtualization against the hypervisor-based solution | Docker | IPERF, HardInfo, STREAM, Bonnie++ | âś“ | âś— | container-based virtualization introduces less overhead compared to hypervisor-based virtualization but could be a bottleneck in storage transaction, report result with limited insight in analysis |
Kuity et al. [10] | 2017 | Performance Evaluation of Container-Based High Performance Computing Ecosystem Using OpenPOWER | Customized Container | HPCC, and IOZone | âś“ | âś“ | container-based solution introduces less overhead compared to hypervisor-based solution in OpenPOWER System, report result with limited insight in analysis |