Abundant examples of complex transaction-oriented networks (TONs) can be found in a variety of disciplines, including information and communication technology, finances, commodity trading, and real estate. A transaction in a TON is executed as a sequ
ence of subtransactions associated with the network nodes, and is committed if every subtransaction is committed. A subtransaction incurs a two-fold overhead on the host node: the fixed transient operational cost and the cost of long-term management (e.g. archiving and support) that potentially grows exponentially with the transaction length. If the overall cost exceeds the node capacity, the node fails and all subtransaction incident to the node, and their parent distributed transactions, are aborted. A TON resilience can be measured in terms of either external workloads or intrinsic node fault rates that cause the TON to partially or fully choke. We demonstrate that under certain conditions, these two measures are equivalent. We further show that the exponential growth of the long-term management costs can be mitigated by adjusting the effective operational cost: in other words, that the future maintenance costs could be absorbed into the transient operational costs.
The power of networks manifests itself in a highly non-linear amplification of a number of effects, and their weakness - in propagation of cascading failures. The potential systemic risk effects can be either exacerbated or mitigated, depending on th
e resilience characteristics of the network. The goals of this paper are to study some characteristics of network amplification and resilience. We simulate random Erdos-Renyi networks and measure amplification by varying node capacity, transaction volume, and expected failure rates. We discover that network throughput scales almost quadratically with respect to the node capacity and that the effects of excessive network load and random and irreparable node faults are equivalent and almost perfectly anticorrelated. This knowledge can be used by capacity planners to determine optimal reliability requirements that maximize the optimal operational regions.
