[Spread-users] throughput degradation

[Martin Decker]

Hello,

In our scenario, we use one sender and 11 receivers. We have a multicast
address defined in our switch and the spread daemon configured with a
single Spread_Segment.

1) test with ./r and ./s
========================

These tests showed, that we have packet misses between 5 and 10%. The
misses seem to be random - not host-dependet and not switch-module
dependent.

2) test with spflooder
======================

As mentioned earlier in this user-group, setting the kernel tuneables of
Solaris 8 to

ndd -set /dev/udp udp_recv_hiwat 32768
ndd -set /dev/udp udp_xmit_hiwat 32768

resulted in zero retrans, which is hard to believe.

flooder: completed multicast of 10000 messages, 1000 bytes each.

Status at host1 V 3.17. 0 (state 1, gstate 1) after 105 seconds :
sent pack:       6      recv pack :   10047     retrans    :       0
u retrans:       0      s retrans :       0     b retrans  :       0
============================
Status at host2 V 3.17. 0 (state 1, gstate 1) after 102 seconds :
sent pack:       6      recv pack :   10047     retrans    :       0
u retrans:       0      s retrans :       0     b retrans  :       0

....

However, what really troubles us is that if we increase the number of
receivers, the time needed to transmit data increases.

With one sender and one receiver, we can transmit 1 GB of data using
spflooder ( time ./bin/spflooder -ro -s 4804 -b 10240 -m 100000) in
about 260 seconds which equals 30.7 MBit/s.

If we increase the number of receivers to 10, the throughput drops to 16.9
MBit/s.

Receiver-Hosts		Übertragungsdauer 		in MBit/s
1				260 sek			30.7 MBit/s
2				276 sek			29.0 MBit/s
3				295 sek			27.1 MBit/s
4				315 sek			25.4 MBit/s
5				326 sek			24.5 MBit/s
6				351 sek			22.8 MBit/s
7				364 sek			22.0 MBit/s
8				449 sek			17.8 MBit/s
9				380 sek			21.1 MBit/s
10				474 sek			16.9 MBit/s

The big question is, how the throughput develops if we will use 20 or 30
receiver hosts.

Can you explain this throughput degradation? Can you explain, why ./r und
./s report misses, but ./spmonitor doesn't?

Best regards,

Martin