From: Marella P-G. <ph...@mo...> - 2005-12-26 17:19:50
|
Hi all,=20 We are observing that the "sendto()" call in TIPC's Hello-World application always returns EAGAIN after a set of messaging-stress tests. This continues to happen well after the stress conditions are withdrawn. Setup:=20 Node1: PowerPC, Linux Kernel=3D 2.6.9,=20 Node2: Dual Xeon (SMP), Linux Kernel=3D 2.6.9,=20 TIPC =3D 1.5.08=20 Communication medium =3D Gigabit Ethernet Sequence leading to problem: The two nodes are brought up and TIPC communication is established. A few (~ 5-6) applications that use TIPC "lightly" are brought up on both nodes. Finally, a test-application that performs several different things (like different message-sizes, multiple subscriptions and inter-node message communication) is continuously spawned and killed on Node 2. The number of subscriptions/bindings never exceed 100 per node. After a few iterations, the test-application blocks on the very first "sendto()" call.=20 Some observations: 1) Only inter-node "sendto()" returns EAGAIN.=20 2) Except for messaging-speed, other system/TIPC resources are not stressed. 3) All communication is connection-less (SOCK_RDM) as in Hello-World 4) Once the problem occurs on Node2, the "RX nacks" count on Node2 keeps on increasing. (see below for a dump) We need help with this problem. With TIPC 1.4.01 this problem is not seen. We would like to know any further information we could supply to help solve this problem.=20 Thanks, Phani Marella PS: Output of tipc-config command after the problem sets in.=20 Link <1.1.8:eth0-1.1.1:eth0> ACTIVE MTU:1500 Priority:10 Tolerance:1500 ms Window:50 packets RX packets:88 fragments:2/1 bundles:0/0 TX packets:2667 fragments:0/0 bundles:826/19803 TX profile sample:1420 packets average:60 octets 0-64:98% -256:1% -1024:0% -4096:0% -16354:0% -32768:0% -66000:0% RX states:9337 probes:4443 naks:1327 defs:0 dups:0 TX states:8883 probes:4440 naks:0 acks:0 dups:0 Congestion bearer:0 link:1981 Send queue max:876 avg:71 |