Linux IPC Profiler/Debugger

[aryan_]

Hi All,

I have an application which contains several C++ modules which run as Linux daemons, and communicate to each other via IPC (specifically, message queues).

For now, the only tool that I know to get useful statistics information about message queues (or other IPCs as well) is "ipcs" command, which lists all active IPC mechanisms in the system.

Is there any free/commercial tool that is capable of providing information about, for example, the number of messages written/read to/from a specific queue per second on a real-time basis, a history info about the messages sent/received, etc ? (To summarize, I want to know exactly what is happening)

I am eager to hear about any tool (if any) that works for other IPC mechanisms as well (shared memories, etc).

I use Ubuntu by the way.

Thanks.

[Rubberman]

As for commercial tools that do this, how much $$ do you have? There are FOSS tools that should do what you want, and if you wish to roll your own, then get the source for ipcs and see how they get the info. Then you can write an application to monitor the queues in "real time". As for a history of the messages sent, you would need to keep some data about each one, which in a busy system would grow to a huge size pretty quickly. I work with manufacturing systems that can easily send a TB of data each day via IPC's of various sorts (local and networked). Some of my clients' systems are processing 10's of thousands of messages per second. The main IPC they use is a propietary message bus (now owned by Applied Materials) that has such monitoring and capabilities. I think a minimum license set for the message bus and related tools is on the order of $100K USD, depending upon how many servers you are going to use. In any case, commercial tools for this sort of purpose tend to be quite expensive, such as HP's Glance, Computer Associates Unicenter, etc.

All that said, you might want to look into Nagios for open source network and system monitoring tools: Nagios - The Industry Standard in IT Infrastructure Monitoring

[aryan_]

I think a minimum license set for the message bus and related tools is on the order of $100K USD, depending upon how many servers you are going to use. In any case, commercial tools for this sort of purpose tend to be quite expensive, such as HP's Glance, Computer Associates Unicenter, etc.

In my case, I do not have networked daemons -- the daemons just use IPC locally. (the order of $100K is too much)

All that said, you might want to look into Nagios for open source network and system monitoring tools

I will take a look at this one since I am assuming that it has a free license.

[Rubberman]

Yes. Nagios is open source and free, with a pretty good user community. They also sell support packages for commercial users.

[aryan_]

Yes. Nagios is open source and free, with a pretty good user community. They also sell support packages for commercial users.

I already took a look at it.

I may be wrong, but Nagios does not seem to provide information (such as, for example, in case of message queues, written/read messages, writing/reading processes, their read/write speeds, etc) about the current state of a certain "local" IPC mechanism.

[Rubberman]

It can be extended to do so. There are a lot of add-ons and plugins for Nagios and it is designed for just such enhancements. The problem is that I also don't find anything that does specifically what you want, yet.