pkrishna at revasystems.com
Tue Jun 22 13:26:59 EDT 2004
Thank you for the prompt response. I read the 2 pager.
Spread maintains only one ring of spread daemons, and the seq# is global
across all groups.
Looking @ Figure 1.4 which plots mesg latency versus number of groups.
If I look at the curve for 10000B packets, the latency is about 2.8ms.
The xmission latency due to pkt over 100Mb link is 0.8ms - what is the
breakup of the remaining 2ms? How much of it is due to CPU (P3 in this
expt) and how much is due to the ordering protocol - i.e., token based
I am just trying to get a gauge of performance expectation if we use a
different setup for example, faster CPUs but continue using 100Mb/s
On Tue, 2004-06-22 at 12:08, Yair Amir wrote:
> P.> Hello Yair,
> P.> Are you using the TOTEM-type (token-passing) algorithm for ensuring
> P.> agreed/safe ordering of messages?
> Yes - the Totem algorithm is the basis for agreed and safe ordering
> during regular operation in the open source version.
> P.> Assuming thats true, then I presume you are doing the following:
> Unfortunately, the description in your e-mail below is *not* a good
> description for Spread. The recent 2-pager we put on the web
> (that will appear as part of Ken Birman book's new edition) can
> answer most of your questions and provide some representative
> performance measurements.
> :) Yair. http://www.cs.jhu.edu/~yairamir
> P.> - For each Spread group, Spread creates a "token ring" and each daemon
> P.> basically keeps track of the TOTEM protocol params (seq#, aru etc) for
> P.> each of the rings it is part of.
> P.> - Each ring has a single seq# which gets incremented upon each message
> P.> getting inserted into the ring. This effectively serializes the messages
> P.> thru a "global fifo" and hence ensures total ordering.
> P.> - the token is the media-access protocol. So, the Token Ring traversal
> P.> time is at least the media access time.
> P.> - Either the last node on the ring for a message discards the message
> P.> (based on zero TTL detection) or maybe the source itself discards the
> P.> message upon getting a match on the src ID.
> P.> The question I have is how many messages can be tagged along with the
> P.> token? Say, I have ten nodes on the ring and say Node 1 initiates the
> P.> token, and then all the nodes along the ring path consume all the
> P.> messages in the token and also transmit a message. What is the max pkt
> P.> size? Of course, it cannot be more than the ethernet-frame-size - what
> P.> if the # of messages being tagged along causes multiple ethernet pkts to
> P.> be sent across, how do you ensure the "group of pkts = single
> P.> token+messages" relationship?
> P.> Do you have results of latencies of agreed ordering w/ respect to
> P.> message traffic, number of members/group. I dont understand how the
> P.> latencies are not much different between agreed and say basic/fifo
> P.> ordering?
> P.> Thanks & Regards,
> P.> Krishna
> P.> _______________________________________________
> P.> Spread-users mailing list
> P.> Spread-users at lists.spread.org
> P.> http://lists.spread.org/mailman/listinfo/spread-users
More information about the Spread-users