Reliable packet delivery within stringent time limits is a key requirement of smart-grid and other industrial communication networks. The Parallel Redundancy Protocol (PRP), used in this context, duplicates packets at the MAC layer over parallel networks and is thus claimed to repair packet loss in 0 ms. However, PRP has several drawbacks: it works only for bridged networks; it requires special hardware; as a layer 2 protocol, it does not have diagnostic tools; it imposes the same MAC addresses to different interfaces, which exacerbates the management difficulty; multicast packet delivery is not natively supported. To address these problems we propose the IP Parallel Redundancy Protocol (IPRP). Unlike PRP, IPRP works in routed networks. It supports UDP applications, unicast or multicast. It is transparent to applications and requires no changes in application software. It is also transparent to the network as IPRP-related packets are regular IP packets; it requires no specific hardware. Operation at the IP layer, possibly with more than two networks makes the design of IPRP very different from PRP. We implemented IPRP in Linux 3.6.11-rt29 using iptables and deployed it on a campus smart grid. We describe our implementation and some performance results.