Dear Danny;
On Oct 25, 2008, at 10:34 PM, Danny Mayer wrote:
Marshall Eubanks wrote:
Sasha,
No, I don't think 10 ms is enough for a stock exchanege at least not
this
particular week given the turbulence on the stock market, but for a
cash
register yes.
(I renamed the thread.)
Given that the financial people are concerned with millisecond
delays,
10 msec
is clearly not enough for some trading applications. See, e.g.,
http://www.usfst.com/currentissue/article.asp?art=275136&issue=281
“a millisecond advantage in trading applications can be worth $100
million a year to a major brokerage firm.”
I read this article and the conclusion I came to was that this has
nothing to do with clock accuracy, it has to do with available
bandwidth
for the critical applications and how fast the network is. The clock
accuracy will not help here, so the 1 millisecond number referenced is
irrelevant.
What is relevant is that the networks people need to be able to do
forensic analysis of packets and application connectivity for which
they
need timestamps to figure out what is happening when and where to
ensure
optimal performance. Nothing in the article talks about the accuracy
needed for that.
No, nor would I expect it to - this was not a technical article.
However, I do not see how you can worry about milliseconds of latency
without having latency measurements with millisecond accuracy or
better. Since they do the one, they need the other.
Much of their use of multicast is driven by such considerations.
True but they are not talking about multicast for time distribution
but
multicast of application data.
True enough. I mentioned their use of multicast (something I have seen
first hand) merely as I thought it was something that people might be
aware of.
Regards
Marshall
Whether they have thought this through or not
is another question.
They may have but so far this is a network performance issue not a
financial issue.
Danny
Regards
Marshall
_______________________________________________
TICTOC mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/tictoc
