Re: investigation into replacing ref. counts with leases

Hi Darin,

The design is different, but not that different. The main
requirement with both the leasing and pinging approach is
that an object knows it's clients.

Pinging is actually more complex because it requires some
periodic (preferably regular) action (ie. pinging the clients)
on the part of each object.

For comparison, DCOM does pinging, but it's handled at the level of the
middleware, not individual objects; the bad side of pinging
is the extra traffic (which could be significant if the
ping period is short, the number of objects large and/or
the bandwidth limited). DCOM does a ping between each
pair of communicating hosts (I guess on the assumption that
a failure to receive a ping is due to a network problem).


>Delivered-To: gnome-components-list gnome org
>From: Darin Adler <darin bentspoon com>
>Subject: Re: investigation into replacing ref. counts with leases
>Cc: Havoc Pennington <hp redhat com>, gnome-components-list gnome org, Michael 
Meeks <michael ximian com>
>To: Martin Baulig <martin home-of-linux org>
>Mime-Version: 1.0 (Apple Message framework v388)
>Content-Transfer-Encoding: 7bit
>On Thursday, July 19, 2001, at 02:20  AM, Martin Baulig wrote:
>> Hmm, what about periodically "pinging" all objects ?
>That does lead to an entirely different design. Instead of leases, every 
>object would have a list of CORBA objects that own it. For owners that are 
>not other CORBA objects, we'd just use a single CORBA object per process 
>to be the proxy owner. An approach like that might work. It's not really 
>ref. counts or leases. It's a third kind of design.
>Someone might want to look into that.
>     -- Darin
>gnome-components-list mailing list
>gnome-components-list gnome org

[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]