The core networking research community is witnessing a trend that many researchers attempt, and some achieve, a high speed packet processing rate using software-based approach. Yet, the extent of performance that can be obtained by deploying only Personal Computers (PCs), or Commercial Off-The-Shelf (COTS) devices, remains as an open-ended question. This is because 1) industrial achievements often involve custom hardware developments that are not counted as COTS devices, and their technical details are not revealed, and 2) research papers do not focus on the goal to answer the question of the best performance achievable using only COTS devices. In this paper, we show our implementation of an 80Gbps PC router using only COTS devices along with our configuration details. We demonstrate achievement of 128B 80Gbps full wire-rate packet forwarding with routing lookups of 500K BGP full-route routing table, using a recent high speed routing lookup technology called Poptrie [1]. We further reveal the conditions that are necessary to construct the 80Gbps high speed PC router by presenting the performance comparisons with and without the specific parameter settings. We conclude with discussion on the promising prospects of software router application in datacenters.

IPv4 addresses are nearly exhausted worldwide. For some time until IPv6 becomes pervasive as an ultimate solution, deployment of Carrier Grade NAT (CGN) devices becomes necessary, especially in the mobile carriers' networks which anticipate a large and growing number of new users. In this context, we tackle the evaluation of the impact of inserting a CGN device in the network, and in conjunction with the mobile network communication delays. We compare the Connection Establishment Rate (CER) with or without a CGN device, also with or without the emulated mobile network communication delays. Against our anticipation, the types of time-varying mobile network delays do not have a significant impact on CER. The effect of the changing delay fades away in the aggregation of many user traffic at the core part of the network, even though the time-varying mobile network communication delays are individually and separately emulated for each user. To the best of our knowledge, this is the first to study the relationship between the CGN performance and the mobile network's communication delays. The result suggests that modeling the aggregate traffic trend (such as the constant delay portion in the network dynamics) is more important rather than emulating each user's traffic separately.