[jgroups-dev] Questions on flow control

--===============7973312146477477429==
Content-Type: multipart/alternative; boundary=047d7bdcaaa6ec049b04e9448660

--047d7bdcaaa6ec049b04e9448660
Content-Type: text/plain; charset=UTF-8

Hi All,

This is Jiadong. I'm new to jgroups, and I have some questions on the flow
control mechanism of jgroups (3.4.0). When conducting the MPerf test on a
3-node cluster with GbE interconnection, I noticed that the default
settings in udp.xml performed poorly under congestion. In 1-to-3
multicasting scenario, default udp.xml delivered an amazing throughput of
120MB/sec/node with 5K-sized message. However, in 3-to-3 multicasting
scenario, default udp.xml only delivered 45MB/sec/node. The throughput
would drop further to single digit if I injected a continues TCP flow onto
one of the nodes as interference.

At first I suspected that the problem was due to sub-optimal MFC
parameters, but after trying several combinations I found out that tuning
the credit parameters could become a dilemma case even without interference
traffic.

It seems to me that lowering the credit pool will increase xmit efficiency
under congestion but will also decrease the best-effort throughput in
a congestion-free environment. With max_credits=50K min_threshold=0.4 and
frag_size=8K, I get a 3-to-3 multicasting throughput of 90MB/sec/node which
is acceptable. However, with such a small credit pool the 1-to-3
multicasting throughput drops to only 48MB/sec/node.

Is it true that the MFC implementation in jgroups 3.4.0 can only handle
congestion statically? If true, then is it viable to improve MFC with some
feedback mechanism (e.g. AMID in TCP)? There seems to be a related
issue JGRP-511 mentioned in previous emails but I didn't find out any
followups on it. Is such feature get ignored due to low worthiness?

Thanks a lot.

Jiadong

--047d7bdcaaa6ec049b04e9448660
Content-Type: text/html; charset=UTF-8
Content-Transfer-Encoding: quoted-printable

<div dir=3D"ltr">Hi All,<div><br></div><div>This is Jiadong. I&#39;m new to=
 jgroups, and I have some questions on the flow control mechanism of jgroup=
s (3.4.0). When conducting the MPerf test on a 3-node cluster with GbE inte=
rconnection, I noticed that the default settings in udp.xml performed poorl=
y under congestion. In 1-to-3 multicasting scenario, default udp.xml delive=
red an amazing throughput of 120MB/sec/node with 5K-sized message. However,=
 in 3-to-3 multicasting scenario, default udp.xml only delivered 45MB/sec/n=
ode. The throughput would drop further to single digit if I injected a cont=
inues TCP flow onto one of the nodes as interference.</div>

<div><br></div><div>At first I suspected that the problem was due to sub-op=
timal MFC parameters, but after trying several combinations I found out tha=
t tuning the credit parameters could become a dilemma case even without int=
erference traffic.</div>

<div><br></div><div>It seems to me that lowering the credit pool will incre=
ase xmit efficiency under=C2=A0congestion=C2=A0but will also decrease the b=
est-effort throughput in a=C2=A0congestion-free environment. With max_credi=
ts=3D50K min_threshold=3D0.4 and frag_size=3D8K, I get a 3-to-3 multicastin=
g throughput of 90MB/sec/node which is acceptable. However, with such a sma=
ll credit pool the 1-to-3 multicasting throughput drops to only 48MB/sec/no=
de.</div>

<div><br></div><div>Is it true that the MFC implementation in jgroups 3.4.0=
 can only handle congestion statically? If true, then is it viable to impro=
ve MFC with some feedback mechanism (e.g. AMID in TCP)? There seems to be a=
 related issue=C2=A0JGRP-511 mentioned in previous emails but I didn&#39;t =
find out any followups on it. Is such feature get ignored due to low worthi=
ness?</div>

<div><br></div><div>Thanks a lot.</div><div><br></div><div>Jiadong</div></d=
iv>

--047d7bdcaaa6ec049b04e9448660--


--===============7973312146477477429==
Content-Type: text/plain; charset="us-ascii"
MIME-Version: 1.0
Content-Transfer-Encoding: 7bit
Content-Disposition: inline

------------------------------------------------------------------------------
October Webinars: Code for Performance
Free Intel webinars can help you accelerate application performance.
Explore tips for MPI, OpenMP, advanced profiling, and more. Get the most from 
the latest Intel processors and coprocessors. See abstracts and register >
http://pubads.g.doubleclick.net/gampad/clk?id=60135991&iu=/4140/ostg.clktrk
--===============7973312146477477429==
Content-Type: text/plain; charset="us-ascii"
MIME-Version: 1.0
Content-Transfer-Encoding: 7bit
Content-Disposition: inline

_______________________________________________
Javagroups-development mailing list


--===============7973312146477477429==--