[cgl_discussion] RE: SCL1.0 discussion continued

Zhuang, Louis louis.zhuang at intel.com
Thu Jul 10 18:10:39 PDT 2003


As Ardelle mentioned in last letter, we'd like to accurate description a
little. :-)  Please see my comments.

  - Louis 
> -----Original Message-----
> From: Fleischer, Julie N
> 1.  System scalability for large number of events.
[Louis] The topic should be "system scalability large number of event
channels"

> - YES - AEM does this with forked event handlers
> - YES - epoll plus a user library to support event handling does this
> 
> 2.  Soft real-time responsiveness for essentially simultaneous events
> - YES - AEM plus RT patch does
> - YES - epoll does as well as AEM
[Louis] Answer here is not clear yet... we need more discussion with
Andre/Ibrahim. But AFAIK, AEM has several hacking in kernel, and _might_
resolve the problem better than epoll. 

> 3.  Events on multiple channels
[Louis] Again, change topic to 'events on multiple types of channels'
for clarification.

> - NEEDS WORK - AEM only supports timer and socket; work needed for
SCTP
> - MAYBE OKAY - supports multiple channels, only read event (this may
be OK)
[Louis] epoll supports accept/read/write/exception etc on socket. Any
others types of channel support depends on implementation, so that's not
a real issue for epoll.

> 4.  Provide Linux kernel with the capability to handle such kind of
events
> => According to Andre Beliveau, this is needed to ensure any
user-space
> solutions work.  Not sure what this means to AEM/epoll.
> 
> 
> **These views are not necessarily those of my employer.**




More information about the cgl_discussion mailing list