Use two empty lines to separate data sets.

[loba.git] / BUGS

diff --git a/BUGS b/BUGS
index ff54efcc2b141bb154acaa153968faa167e7cad9..b1cc865db3ea925fcf8f4a610153f738a165864e 100644 (file)
--- a/BUGS
+++ b/BUGS
@@ -1,3 +1,43 @@
-./loba cluster1000.xml -N64 -L100 -i100 -a fair -T hcube
-=> leads to deadlock (with stable)
+Comment expliquer ces différences entre SG 3.5 et SG svn ?

 
 

+$ ./loba platform.xml -Tline -abest -L500 -t1800 -c1e8,0 -C1e7,0 -M0 2>&1 --log=comm.thres:debug -s100 -b | grep LOAD
+[Bourassa 1031.097913] [comm/DEBUG] send LOAD: 366.211 to Fafard_data
+[Bourassa 1306.997159] [comm/DEBUG] send LOAD: 20.0806 to Fafard_data
+[Bourassa 1541.486345] [comm/DEBUG] send LOAD: 4.74548 to Fafard_data
+[Bourassa 1766.189415] [comm/DEBUG] send LOAD: 3.09753 to Fafard_data
+[Fafard 2579.229566] [comm/DEBUG] received LOAD: 366.211 from Bourassa
+[Fafard 2605.989948] [comm/DEBUG] received LOAD: 20.0806 from Bourassa
+[Fafard 2612.318155] [comm/DEBUG] received LOAD: 4.74548 from Bourassa
+[Fafard 2616.450666] [comm/DEBUG] received LOAD: 3.09753 from Bourassa
+
+$ ./loba-dev platform_dev.xml -Tline -abest -L500 -t1800 -c1e8,0 -C1e7,0 -M0 2>&1 --log=comm.thres:debug -s100 -b | grep LOAD
+[Bourassa 1031.097913] [comm/DEBUG] send LOAD: 366.211 to Fafard_data
+[Bourassa 1306.997159] [comm/DEBUG] send LOAD: 20.0806 to Fafard_data
+[Fafard 1519.035900] [comm/DEBUG] received LOAD: 366.211 from Bourassa
+[Fafard 1519.035900] [comm/DEBUG] send LOAD: 282.074 to Ginette_data
+[Bourassa 1541.486345] [comm/DEBUG] send LOAD: 4.74548 to Fafard_data
+[Fafard 1629.312931] [comm/DEBUG] received LOAD: 20.0806 from Bourassa
+[Fafard 1629.312931] [comm/DEBUG] received LOAD: 4.74548 from Bourassa
+[Fafard 1629.312931] [comm/DEBUG] send LOAD: 6.19507 to Ginette_data
+[Bourassa 1766.189415] [comm/DEBUG] send LOAD: 3.09753 to Fafard_data
+[Fafard 1898.705676] [comm/DEBUG] received LOAD: 3.09753 from Bourassa
+[Ginette 1932.076243] [comm/DEBUG] received LOAD: 282.074 from Fafard
+[Ginette 1940.343540] [comm/DEBUG] received LOAD: 6.19507 from Fafard
+
+========================================================================
+Il semblerait qu'il y ait un bug dans SG 3.5, et qu'on ne puisse pas
+utiliser MSG_comm_waitany() pour l'émetteur *et* le récepteur sans
+risquer d'interblocage.
+
+Le problème devrait être contourné correctement depuis le commit
+cd6b253 Use MSG_comm_waitall for communicator::flush(true).
+
+========================================================================
+Avec SG 3.5, les communications doivent être détruites dès que
+possible avec MSG_comm_destroy().  Si ce n'est pas fait, la simulation
+peut être extrêmement ralentie.
+
+Le problème devrait être contourné correctement depuis le commit
+404a8d5 Do not call flush automatically in communcator::send...
+
+========================================================================