Still Alive: Extending Keep-Alive Intervals in P2P Overlay Networks

Abstract

Abstract-Nodes within existing P2P networks typically exchange periodic keep-alive messages in order to maintain network connections between neighbours. This paper investigates a number of algorithms which allow each individual connections to extend the interval between successive keep-alive messages based upon the likelihood that a corresponding node will remain in the system.Several studies have shown that older peers are more likely to remain in the network longer than their short-lived counterparts. Therefore using the distribution of peer session times and the current age of peers as key attributes, we propose three algorithms that increase the interval between successive keepalive messages as nodes become more reliable.By prioritising keep-alive messages to nodes that are more likely to fail, our algorithms reduce the expected delay between failures occurring and their subsequent detection. Failed connections can incur expensive lookup timeouts and increases the network's vulnerability to partitioning. We extensively analyse the properties of these algorithms and compare them to the standard periodic keep-alive mechanism using simulations based upon measured network data.

Adapting to NAT timeout values in P2P overlay networks

“…Cost and mean failure detection delay have been used as performance metrics other evaluations of failure detection algorithms including [7], [8], [10].…”

“…We would also like to extend our previous work presented in [7] to improve failure detection within structured networks. As structured networks must maintain the topology of the overlay they present a different constraints to the unstructured networks examined previously.…”

“…Our previous work in [7], introduced three algorithms that determined the required frequency of keep-alive messages by using the current uptime to predict the remaining uptime of nodes. By prioritising keep-alive messages to nodes that are more likely to fail, the expected delay between failures occurring and their subsequent detection can be reduced.…”

“…If a node leaves the network it does not respond to keep-alive messages, these unacknowledged messages are typically re-sent multiple times at short intervals to minimise the risk of false positives where keep-alive messages have been somehow lost. As ungraceful Figure 1a as shown in [7] illustrates node i arriving in the network at time a i and departing the network at time d i . A node's session time S i , the amount of time node i spends in the network is given by S i = d i − a i .…”

“…While application developers may want to reduce the failure detection delay as much as possible, using the SKA approach this can only be done at the cost of sending more keep-alive messages. While other techniques [7], [8], [9], [10] improve upon the performance of the SKA approach the fundamental trade-off remains the same; extending the keepalive period increases the incurred failure detection delay while reducing the cost in terms of bandwidth.…”