Bug#914577: RFP: dropwatch -- A tool for detecting and diagnosing packets being dropped

[Dean Hamstead <dean@fragfest.com.au>]

Package: wnpp
Severity: wishlist

* Package name    : dropwatch
  Version         : 1.5
  Upstream Author : Neil Horman <nhorman@redhat.com>
* URL             : https://github.com/nhorman/dropwatch
* License         : GPL
  Programming Lang: C
  Description     : A tool for detecting and diagnosing packets being dropped

What is Dropwatch? Dropwatch is a project I started in an effort to improve the ability for developers and system administrator to diagnose problems in the Linux Networking stack, specifically in our ability to diagnose where packets are getting dropped. From my probing, I've come to the conclusion that there are four main shortcommings in our current environment:

1.    Consolidation, or lack thereof. Currently, if you would like to check on the status of dropped packets in the kernel, you need to check at least 3 places, and possibly more: The /proc/net/snmp file, the netstat utility, the tc utility, and ethool. This project aims to consolidate several of those checks into one tool, making it easier for a sysadmin or developer to detect lost packets

2.    Clarity of information. Dropped packets are not obvious. a sysadmin needs to be intimately familiar with each of the above tools to understand which events or statistics correlate to a dropped packet and which do not. While that is often self evident, it is also often not. Dropwatch aims to improve that clarity

3.    Ambiguity. Even when a dropped packet is detected, the causes for those dropped packets are not always clear. Does a UDPInError mean the application receive buffer was full, or does it mean its checksum was bad? Dropwatch attempts to disambiguate the causes for dropped packets.

4.    Performance. Utilties can be written to aggregate the data in the various other utilities to solve some of these problems, but such solutions require periodic polling of several interfaces, which is far from optimal, especially when lost packets are rare. This solution improves on the performance aspect by implementing a kernel feature which allows asyncronous notification of dropped packets when they happen.