Understanding the network level performance of virtualization solutions

Arsene, Adrian; Lopez-Pacheco, Dino; Urvoy-Keller, Guillaume

doi:10.1109/cloudnet.2012.6483645

Cited by 4 publications

(3 citation statements)

References 7 publications

(7 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Thus, despite the per-packet extra-delay introduced by virtualization, as reported in [10], the response times of the virtualized case are very close to the one of the native case. We believe that the explanation behind this observation lies in the modus operandi of the TCP protocol.…”

Section: B Virtualization Overheadsupporting

confidence: 66%

“…The virtualization overhead mostly stems from the transfer of the packets from the guest OS to the hypervisor (back-end guest in the case of Xen -see Figure 10). If the guest is delivering one packet at a time at a constant rate, as it is the case with the CBR UDP transfers considered in [10], the overhead results in a higher jitter. However, for the case of TCP, as packets are delivered in bursts, and the buffers are never empty in the steady-state, the transfer cost from the VM to the hypervisor is amortized (as soon as a time slice is assigned to a VM, it will send all its accumulated data if possible).…”

Section: B Virtualization Overheadmentioning

confidence: 97%

“…Following the success of virtualization, several researches have been done to understand the performance of virtualization and/or to propose new features [2], [11], [1], [10], [12].…”

Section: Related Workmentioning

confidence: 99%

See 2 more Smart Citations

Networking in a virtualized environment: The TCP case

Ha¹,

Lopez-Pacheco²,

Urvoy-Keller³

2013

2013 IEEE 2nd International Conference on Cloud Networking (CloudNet)

Self Cite

View full text Add to dashboard Cite

Virtualization is heavily used in modern private and public data centers. By enabling the consolidation of many virtual machines (VMs) in a physical server, virtualization blurs the frontier between networking and system. We have investigated in a previous work the various prices due to virtualization when a packet is moved from one VM to the physical NIC, by focusing on a pure UDP scenario and multiple competing VMs. We estimated how the inter-packet delay moves away from the optimal point as the number of VMs increases. In the present article we report the impact of virtualization over a whole data stream controlled by the TCP protocol. Our findings reveal that the impact of the additional inter-packet delay is negligible on the performance of TCP. Indeed, virtualized systems have similar goodput levels to the ones of natives systems. Moreover, a less expected result show that virtualized systems improve the fairness between flows, even when such flows are streamed from a same VM. The bad consequences of the virtualization concern mainly bugs in the implementation of the networking function, which can potentially introduce spurious retransmissions and information leaks.

show abstract

Section: B Virtualization Overheadsupporting

confidence: 66%

Section: B Virtualization Overheadmentioning

confidence: 97%