[Barrelfish-users] about remote communication on barrelfish

Andrew Baumann Andrew.Baumann at microsoft.com
Mon Mar 11 19:58:58 CET 2013

Hi Xiaowen,

The monitor is typically involved in setting up bindings for inter-core communication, but does not have to be on the data path for each message. For UMP, the monitor helps setup a region of shared memory between the dispatchers, but the actual messages travel directly between dispatchers using this shared memory.

I'm not sure what the Tile's message architecture looks like, but I would hope it has a similar structure ... having the monitor on the data path for each message would be a huge overhead. If you can share more about the hardware mechanisms available to you, perhaps we can suggest in more detail how to implement this in Barrelfish.


From: Xiaowen Wang [mailto:xiaowenw at kth.se]
Sent: Monday, 11 March 2013 11:54
To: barrelfish-users at lists.inf.ethz.ch
Subject: [Barrelfish-users] about remote communication on barrelfish

Hi, all,

I am porting Barrelfish onto TilePro architecture, now I am using TilePro's own network structure to implement core-to-core communication. But first of all I want to know some Barrelfish requirements about inter-core communication.

As far as I know, monitor is responsible for inter-core communication in Barrelfish, does that mean any remote message should pass through the local monitor to remote monitor and then is sent to according dispatcher by remote monitor?

For example, if one dispatcher on core 0 sends UMP message to another dispatcher on core 1, so this case is totally wrong? By the way, I am not using UMP, but in my case, I saw some dispatcher (not monitor) tries to send a remote message.

Best regards,
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://lists.inf.ethz.ch/pipermail/barrelfish-users/attachments/20130311/1aaa2353/attachment.html 

More information about the Barrelfish-users mailing list