Known Issues
This page lists known issues with PL-VINI as a whole and its various software packages.
XORP
The "fail a link" experiment doesn't quite work as described in the VINI paper. XORP doesn't fall back to the old path after the failed link is brought back up.
This appears to be due to a bug that has snuck into XORP. VINI currently uses XORP 1.3, while the experiments in the VINI paper used a pre-1.3 build of XORP from the CVS tree.
Quagga
Everything seems OK?
Click
Sometimes a new connection through the NAT elements will not succeed.
The problem is that the NAT can choose a local port that is already in use. This should be fixed so that only free ports are chosen.
UML
When forwarding packets in UML, a number of messages like "send_sock sending to fd 5 Resource temporarily unavailable" scroll up the screen
This is an issue with the uml_switch. These messages are associated with packets dropped by uml_switch. We are evaluating patches that claim to fix the problem.
OpenVPN
The OpenVPN server does not respond to pings.
Even though the OpenVPN server has an IP address (e.g., 10.1.89.5) there is no actual interface with this address. The Click configuration could be changed to respond to pings on the OpenVPN server's behalf if this behavior is desirable.
PL-VINI
Throughput is low.
The SIGCOMM paper measured about 200Mb/s end-to-end TCP throughput on PL-VINI. While this should be sufficient for running an experiment or demonstrating a small-scale prototype, it may be too low for PL-VINI to support a long-running, widely used network service. Click appears to be a bottleneck: by replacing Click with packet forwarder hand-coded in C, we have measured over 500Mb/s TCP throughput. We are currently trying to understand how to optimize our Click configuration for better performance.