[Spread-users] Java API Performance
Mike Perik
michaelperik at yahoo.com
Wed Apr 23 16:24:04 EDT 2008
I'd make it optional. I added a method to set a flag to true so that after the socket creation TcpNoDelay would be set on the socket object.
Another issue I've seen is when the publisher sends a lot of messages fast. I think this is talked about in some of the documents available from the web site. If I throttle it a little (sleep for at least a millisecond) the latency doesn't suffer. Without throttling, the latency jumps to over a 1/4 of a second.
Mike
--- On Wed, 4/23/08, John Schultz <jschultz at spreadconcepts.com> wrote:
> From: John Schultz <jschultz at spreadconcepts.com>
> Subject: Re: [Spread-users] Java API Performance
> To: "Mike Perik" <michaelperik at yahoo.com>
> Cc: spread-users at lists.spread.org
> Date: Wednesday, April 23, 2008, 1:36 PM
> Sounds like a good idea to include in the code to me.
>
> ---
> John Schultz
> Spread Concepts
> Phn: 443 838 2200
>
> On Wed, 23 Apr 2008, Mike Perik wrote:
>
> > Found something, I set TcpNoDelay to 'true' on
> the socket in the SpreadConnection class and my times
> dropped to ~1ms
> >
> > Any thoughts on this?
> >
> > Thanks,
> > Mike
> >
> >
> > --- On Wed, 4/23/08, Mike Perik
> <michaelperik at yahoo.com> wrote:
> >
> >> From: Mike Perik <michaelperik at yahoo.com>
> >> Subject: Re: [Spread-users] Java API Performance
> >> To: spread-users at lists.spread.org
> >> Date: Wednesday, April 23, 2008, 9:48 AM
> >> Anyone have any idea where this latency is coming
> from? It
> >> makes the use of the Java API useless.
> >>
> >> Is there anything I can do to see when the Spread
> daemon
> >> gets the message? I've added milliseconds to
> the
> >> logging done by the daemon but I'm finding it
> difficult
> >> to track the receipt and sending of the
> application
> >> messages.
> >>
> >> Has anyone done a JNI version of the C API?
> >>
> >> Thanks,
> >> Mike
> >>
> >>
> >> --- On Mon, 4/21/08, Mike Perik
> >> <michaelperik at yahoo.com> wrote:
> >>
> >>> From: Mike Perik
> <michaelperik at yahoo.com>
> >>> Subject: Re: [Spread-users] Java API
> Performance
> >>> To: spread-users at lists.spread.org
> >>> Date: Monday, April 21, 2008, 3:52 PM
> >>> I run a second instance of the process that
> only
> >> receives
> >>> and it has the same latency issue. The sender
> >> doesn't
> >>> have to listen but I am doing it to check the
> latency
> >> there
> >>> as well.
> >>>
> >>> I've removed the sender doing a receive
> and the
> >> latency
> >>> dropped to approx 1 ms.
> >>>
> >>> Why would the latency increase when I attempt
> to
> >> receive?
> >>> If I don't do a receive (I also don't
> join the
> >>> group the message is sent to) and just blast
> the
> >> messages
> >>> out the latency is high also. If I sleep or
> 1 ms
> >> after
> >>> doing the send the latency drops down to ~1 ms
> again.
> >>>
> >>> I also did a test where I sent the first 2500
> messages
> >>> without joining the "LATENCY" group
> and the
> >> times
> >>> were ~1 ms. On the 2500th message I had the
> sender
> >> joining
> >>> the "LATENCY" group and the latency
> number
> >> jumped
> >>> from between 2ms - 41 ms. Seems to increment
> by 2 ms
> >> per
> >>> order.
> >>>
> >>> Recver Latency: 2
> >>> Recver Latency: 41
> >>> Recver Latency: 39
> >>> Recver Latency: 37
> >>> Recver Latency: 35
> >>> Recver Latency: 33
> >>> Recver Latency: 31
> >>> Recver Latency: 29
> >>> Recver Latency: 27
> >>> Recver Latency: 25
> >>> Recver Latency: 23
> >>> Recver Latency: 21
> >>> Recver Latency: 19
> >>> Recver Latency: 17
> >>> Recver Latency: 16
> >>> Recver Latency: 14
> >>> Recver Latency: 12
> >>> Recver Latency: 10
> >>> Recver Latency: 8
> >>> Recver Latency: 6
> >>> Recver Latency: 4
> >>> Recver Latency: 2
> >>> Recver Latency: 41
> >>> Recver Latency: 39
> >>> Recver Latency: 37
> >>> Recver Latency: 35
> >>> Recver Latency: 33
> >>> Recver Latency: 31
> >>> Recver Latency: 29
> >>> Recver Latency: 27
> >>> Recver Latency: 25
> >>> Recver Latency: 24
> >>> Recver Latency: 22
> >>> Recver Latency: 20
> >>> Recver Latency: 18
> >>> Recver Latency: 16
> >>> Recver Latency: 14
> >>> Recver Latency: 12
> >>> Recver Latency: 10
> >>> Recver Latency: 8
> >>> Recver Latency: 6
> >>> Recver Latency: 4
> >>> Recver Latency: 2
> >>> Recver Latency: 41
> >>> Recver Latency: 39
> >>> Recver Latency: 37
> >>>
> >>> Any ideas on how I can be a subscriber and a
> publisher
> >> on
> >>> the same group without incurring this huge
> time
> >> penalty?
> >>>
> >>> Am I doing something wrong here?
> >>>
> >>> Mike
> >>>
> >>>
> >>> --- On Mon, 4/21/08, John Keating
> >>> <jkeating at ca.afilias.info> wrote:
> >>>
> >>>> From: John Keating
> >> <jkeating at ca.afilias.info>
> >>>> Subject: Re: [Spread-users] Java API
> Performance
> >>>> To: michaelperik at yahoo.com
> >>>> Cc: spread-users at lists.spread.org
> >>>> Date: Monday, April 21, 2008, 2:51 PM
> >>>> When you're the sender, you're
> also
> >> calling
> >>>> receiveMessage()...should you?
> >>>>
> >>>> Mike Perik wrote:
> >>>>> Has anyone done any basic performance
> tests
> >> with
> >>> the
> >>>> Java API?
> >>>>> I'm seeing what I would consider
> lengthy
> >>> sending
> >>>> times.
> >>>>>
> >>>>> My test involves a process sending
> 5000
> >> messages.
> >>> The
> >>>> message is the current time with the
> granularity
> >> to
> >>>> milliseconds. The sending process sends a
> >> message and
> >>> then
> >>>> receives the message. When it receives
> the
> >> message is
> >>> takes
> >>>> time it received it and subtracts the
> sending
> >> time.
> >>> I'm
> >>>> seeing a round trip time of 39-41
> milliseconds.
> >> I did
> >>> the
> >>>> same test in C and the round trip time is
> more
> >> like 1
> >>> ms.
> >>>>>
> >>>>> Maybe I'm doing something
> incorrectly,
> >> if so
> >>>> I'm not sure what it is and would be
> >> interested in
> >>> any
> >>>> help you can give me.
> >>>>>
> >>>>> Below is the code I'm using.
> >>>>>
> >>>>> Thank you for your help,
> >>>>> Mike
> >>>>>
> >>>>>
> >>>>> import spread.*;
> >>>>> import java.net.*;
> >>>>>
> >>>>> public class spreadlatency
> >>>>> {
> >>>>> SpreadConnection _sc;
> >>>>> boolean _sender = false;
> >>>>> MessageFactory _mf;
> >>>>> static public void main(String
> args[])
> >> throws
> >>>> Exception
> >>>>> {
> >>>>> SpreadConnection sc = new
> >>> SpreadConnection();
> >>>>> spreadlatency sl = new
> >> spreadlatency(sc,
> >>>> args.length == 0 ? true : false);
> >>>>>
> >> sc.connect(InetAddress.getLocalHost(),
> >>> 4803,
> >>>> null, false, false);
> >>>>> sl.run();
> >>>>> }
> >>>>>
> >>>>> public
> spreadlatency(SpreadConnection
> >> sc,
> >>> boolean
> >>>> sender) throws Exception
> >>>>> {
> >>>>> _sc = sc;
> >>>>> _sender = sender;
> >>>>> }
> >>>>>
> >>>>> public void run()
> >>>>> {
> >>>>> SpreadGroup sg = new
> SpreadGroup();
> >>>>> try {
> >>>>> sg.join(_sc,
> >> "LATENCY");
> >>>>> }
> >>>>> catch (Exception ex)
> >>>>> {
> >>>>> ex.printStackTrace();
> >>>>> }
> >>>>> if (_sender == true)
> >>>>> {
> >>>>> SpreadMessage sm = new
> >>> SpreadMessage();
> >>>>> short y = 1000;
> >>>>> sm.setType(y);
> >>>>>
> >> sm.addGroup("LATENCY");
> >>>>> int x = 0;
> >>>>> while (x++ < 5000)
> >>>>> {
> >>>>> //
> String st
> >> =
> >>>> "";
> >>>>> long t =
> >>> System.currentTimeMillis();
> >>>>> // st
> += t;
> >>>>> //
> >>>> sm.setData(st.getBytes());
> >>>>> try {
> >>>>> sm.setObject(new
> >> Long(t));
> >>>>> _sc.multicast(sm);
> >>>>> receiveMessage();
> >>>>> }
> >>>>> catch (Exception ex)
> >>>>> {
> >>>>>
> ex.printStackTrace();
> >>>>> }
> >>>>>
> >>>>> }
> >>>>> }
> >>>>> else
> >>>>> {
> >>>>> int x = 0;
> >>>>> while (x++ < 5000)
> >>>>> {
> >>>>> receiveMessage();
> >>>>> }
> >>>>> }
> >>>>> }
> >>>>>
> >>>>> void receiveMessage()
> >>>>> {
> >>>>> try {
> >>>>> SpreadMessage sm =
> >> _sc.receive();
> >>>>> short type = sm.getType();
> >>>>> if (type == 1000)
> >>>>> {
> >>>>> //
> String s =
> >> new
> >>>> String(sm.getData());
> >>>>> // Long
> l =
> >>>> Long.valueOf(s);
> >>>>> Long l =
> >> (Long)sm.getObject();
> >>>>> if (_sender == true)
> >>>>> {
> >>>>>
> >>> System.out.println("Sender
> >>>> Latency: " +
> (System.currentTimeMillis() -
> >>>> l.longValue()));
> >>>>> }
> >>>>> else
> >>>>> {
> >>>>>
> >>> System.out.println("Recver
> >>>> Latency: " +
> (System.currentTimeMillis() -
> >>>> l.longValue()));
> >>>>> }
> >>>>> }
> >>>>> }
> >>>>> catch (Exception ex)
> >>>>> {
> >>>>> ex.printStackTrace();
> >>>>> }
> >>>>> }
> >>>>> }
> >>>>>
> >>>>>
> >>>>>
> >>>>>
> >>>>>
> >>>>>
> >>>>
> >>>
> >>
> ____________________________________________________________________________________
> >>>>> Be a better friend, newshound, and
> >>>>> know-it-all with Yahoo! Mobile. Try
> it now.
> >>
> >>>>
> >>>
> >>
> http://mobile.yahoo.com/;_ylt=Ahu06i62sR8HDtDypao8Wcj9tAcJ
> >>>>>
> >>>>>
> >> _______________________________________________
> >>>>> Spread-users mailing list
> >>>>> Spread-users at lists.spread.org
> >>>>>
> >>>
> http://lists.spread.org/mailman/listinfo/spread-users
> >>>>>
> >>>
> >>>
> >>>
> >>>
> >>
> ____________________________________________________________________________________
> >>> Be a better friend, newshound, and
> >>> know-it-all with Yahoo! Mobile. Try it now.
> >>>
> >>
> http://mobile.yahoo.com/;_ylt=Ahu06i62sR8HDtDypao8Wcj9tAcJ
> >>>
> >>>
> _______________________________________________
> >>> Spread-users mailing list
> >>> Spread-users at lists.spread.org
> >>>
> http://lists.spread.org/mailman/listinfo/spread-users
> >>
> >>
> >>
> >>
> ____________________________________________________________________________________
> >> Be a better friend, newshound, and
> >> know-it-all with Yahoo! Mobile. Try it now.
> >>
> http://mobile.yahoo.com/;_ylt=Ahu06i62sR8HDtDypao8Wcj9tAcJ
> >>
> >> _______________________________________________
> >> Spread-users mailing list
> >> Spread-users at lists.spread.org
> >>
> http://lists.spread.org/mailman/listinfo/spread-users
> >
> >
> >
> ____________________________________________________________________________________
> > Be a better friend, newshound, and
> > know-it-all with Yahoo! Mobile. Try it now.
> http://mobile.yahoo.com/;_ylt=Ahu06i62sR8HDtDypao8Wcj9tAcJ
> >
> > _______________________________________________
> > Spread-users mailing list
> > Spread-users at lists.spread.org
> > http://lists.spread.org/mailman/listinfo/spread-users
> >
____________________________________________________________________________________
Be a better friend, newshound, and
know-it-all with Yahoo! Mobile. Try it now. http://mobile.yahoo.com/;_ylt=Ahu06i62sR8HDtDypao8Wcj9tAcJ
More information about the Spread-users
mailing list