Comments for Steve Huston's Networked Programming Blog

Comment on Review of AMQP Face-to-Face Public Meeting, April 1, 2009 by stevehuston

stevehuston — Fri, 20 Nov 2009 18:52:00 +0000

Good question, Eric. You’re right that AMQP is a layer above ACE. One could build an AMQP implementation using ACE.

ACE: broad C++ OS middleware – it covers much more than comms, as it has facilities for sockets, threads, containers, memory pools, synch primitives, etc. In terms of comparing with AMQP for communications, ACE is great for when you need TCP-level flexibility or you want direct control of the connection. ACE would also be more appropriate if you must interact with a peer whose software you don’t control; i.e., the protocol is defined beforehand and you have no control over that. ACE also has a big advantage in platform and compiler coverage at this time.

AMQP: all about business messaging (think MSMQ or IBM MQ Series). If the problem you are solving involves exchanging messages that are not necessarily 1:1 peer-to-peer, need reliable queuing, can be persistent across failures, or that involve transactions, AMQP is better suited. Again, you could build an AMQP with ACE, but why would you? There are multiple AMQP implementations available to use already, in various languages.

That said, ACE and AMQP are not mutually exclusive. You can, for example, use AMQP for messaging and ACE to handle threads or work queuing or portable shared memory. It would also be possible, for example, to develop an ACE Reactor implementation that handles AMQP messages.

Comment on Review of AMQP Face-to-Face Public Meeting, April 1, 2009 by Eric Strennen

Eric Strennen — Fri, 20 Nov 2009 16:13:50 +0000

I am just starting to get up to speed on AMQP. I am the principal developer of a scientific application that has disparate programs that communicate via sockets using the ACE framework (on a Windows platform). To me it appears as though AMQP is a layer abstracted above ACE. Would you be able to briefly describe the strengths/weaknesses through a comparison of these two? Thanks.

Comment on Revised ACE_Dev_Poll_Reactor Fixes Multithread Issues (and more!) on Linux by stevehuston

stevehuston — Fri, 25 Sep 2009 21:21:12 +0000

Ok, thanks Herb… there’s clearly something going wrong somewhere. I’ve been very busy with customers, but if you and Sergio can get together and work out a fix, I’ll try to help apply it when I have free time. Or, if a customer raises this issue it’ll get attention quickly.

Comment on Revised ACE_Dev_Poll_Reactor Fixes Multithread Issues (and more!) on Linux by Herb Gillman

Herb Gillman — Fri, 25 Sep 2009 21:06:58 +0000

I too am seeing this odd behavior with the dev_poll reactor. Its bizzare. When I start my app, the top CPU utilization goes to > 100% even though nothing really is running. When I run under the debugger, all the Reactor thread’s from my ThreadPool exit and everything appears to run on one thread on one CPU, the top CPU utilization on startup is 0 (as expected).

I am running on AMD 3Ghz quad core boxes, one with Linux RHEL4.5 and one with Linux RHEL5.3
ACE version is 5.7.3-2 (downloaded and built from http://dist.bonsai.com/ken/ace_tao_rpm/SRC/)

Comment on Revised ACE_Dev_Poll_Reactor Fixes Multithread Issues (and more!) on Linux by Sergio Ruiz

Sergio Ruiz — Thu, 02 Jul 2009 18:57:38 +0000

Hi Steve,
I’ve sent you the output from scall() and the one from my test app using ACE_TRACE. The trace shows a large number of sys calls to futex, probably related to the reactor’s token.
Thanks,
Sergio

Comment on When is it ok to use ACE Proactor on Linux? by stevehuston

stevehuston — Thu, 02 Jul 2009 14:32:39 +0000

Yes, Gokul, you are right. Solaris is better for two reasons: it doesn’t suffer from the same “silently convert to synch” problem as Linux does, and there is a separate Proactor implementation on Solaris that takes advantage of Solaris’s features better.

Comment on When is it ok to use ACE Proactor on Linux? by Gokul

Gokul — Thu, 02 Jul 2009 07:29:41 +0000

Is it safe to user ACE Proactor in Solaris? I think there is a separate implementation in ACE to take advantage of the Asynchronous I/O implementation in Solaris.

Comment on Revised ACE_Dev_Poll_Reactor Fixes Multithread Issues (and more!) on Linux by stevehuston

stevehuston — Tue, 30 Jun 2009 14:46:30 +0000

Hey Sergio… right, I saw your problem report on ace_users. This hasn’t come up anywhere else, and I haven’t had free time to examine the test program you sent. If you can help collect some profiling info to help narrow down the search that may help.

Comment on Why Apache Qpid’s C++ Build is Switching to CMake by stevehuston

stevehuston — Tue, 30 Jun 2009 14:42:01 +0000

Thanks very much – glad to have you participating here.

Comment on Revised ACE_Dev_Poll_Reactor Fixes Multithread Issues (and more!) on Linux by Sergio Ruiz

Sergio Ruiz — Fri, 26 Jun 2009 02:27:03 +0000

Hi Steve,
After I upgraded my server app from ACE 5.6.2 to 5.7, I noticed a high CPU utlization > 100%. My server uses multiple worker threads to run the ACE_Dev_Poll_Reactor. I got this behavior on CentOS 4.x and 5.x.
My app works fine if I revert back to v5.6.2 or if I run it with TP_Reactor.
Has anybody reported this issue yet?
Any ideas what may be causing it?
Thanks,
Sergio.