Electing a network leader
10 views
Skip to first unread message
Mickey
May 5, 2010, 10:08:44 AM5/5/10
to nextgen_engg
There can be several approaches... what do we use in real networks in
such situations?
Describe a technique to identify a "leader" among a group of 10
identical servers that are all connected to every other server. There
are no prior distinguishing characteristics of any of them and the
same program to identify the leader starts running on all of them at
the same time.
Thank you,
Jyoti
such situations?
Describe a technique to identify a "leader" among a group of 10
identical servers that are all connected to every other server. There
are no prior distinguishing characteristics of any of them and the
same program to identify the leader starts running on all of them at
the same time.
Thank you,
Jyoti
SUDHEER DURUSOJU
May 5, 2010, 11:23:51 AM5/5/10
to nextge...@googlegroups.com
I will base my program on the Network delays in communication. Thought the nodes mentioned are connected the medium of connection and distance between them can make a big difference.
I will send a ping request to all of the nodes and take the average of "Average turn around time" for all nodes.
I will make the node which had lowest average as the leader.
-Sudheer
I will send a ping request to all of the nodes and take the average of "Average turn around time" for all nodes.
I will make the node which had lowest average as the leader.
-Sudheer
Mickey
May 5, 2010, 1:31:36 PM5/5/10
to nextgen_engg
This is an interesting idea. But I think this is not a popular choice
because of scalability?
Regards,
Jyoti
On May 5, 8:23 pm, SUDHEER DURUSOJU <sdurus...@gmail.com> wrote:
> I will base my program on the Network delays in communication. Thought the
> nodes mentioned are connected the medium of connection and distance between
> them can make a big difference.
>
> I will send a ping request to all of the nodes and take the average of
> "Average turn around time" for all nodes.
>
> I will make the node which had lowest average as the leader.
>
> -Sudheer
>
because of scalability?
Regards,
Jyoti
On May 5, 8:23 pm, SUDHEER DURUSOJU <sdurus...@gmail.com> wrote:
> I will base my program on the Network delays in communication. Thought the
> nodes mentioned are connected the medium of connection and distance between
> them can make a big difference.
>
> I will send a ping request to all of the nodes and take the average of
> "Average turn around time" for all nodes.
>
> I will make the node which had lowest average as the leader.
>
> -Sudheer
>
Mickey
May 5, 2010, 1:43:26 PM5/5/10
to nextgen_engg
And the fact that "the same program to identify the leader starts
running on all of them at the same time" makes this algorithm
difficult to use.
Regards,
Jyoti
running on all of them at the same time" makes this algorithm
difficult to use.
Regards,
Jyoti
SUDHEER DURUSOJU
May 5, 2010, 1:46:40 PM5/5/10
to nextge...@googlegroups.com
Agreed :) Looking for more ideas ...
-Sudheer
-Sudheer
Akhil Bhiwal
May 15, 2010, 11:45:54 AM5/15/10
to nextgen_engg
@Jyoti: As there is no response since the last ten days, please post
the solution to this problem.
the solution to this problem.
Mickey
May 15, 2010, 10:09:01 PM5/15/10
to nextgen_engg
There is no perfect solution to it. I was just looking for ideas.
One possible approach (which is actually used in some networks) is to
elect the mode with lowest/highest ip/number as the leader.
Usually the most popular choice is to have all of the nodes send out a
message (with timestamp) at random intervals and the one who was first
(earliest timestamp) is chosen as the leader. The process is repeated
in case of a collision. The essence being that there has to be an
element of randomness when otherwise all the nodes are identical in
performance.
Regards,
Jyoti
One possible approach (which is actually used in some networks) is to
elect the mode with lowest/highest ip/number as the leader.
Usually the most popular choice is to have all of the nodes send out a
message (with timestamp) at random intervals and the one who was first
(earliest timestamp) is chosen as the leader. The process is repeated
in case of a collision. The essence being that there has to be an
element of randomness when otherwise all the nodes are identical in
performance.
Regards,
Jyoti
Reply all
Reply to author
Forward
0 new messages