- **Comment**:
Sending message with TIPC_ADDR_MCAST addr_type when snd_type is equal to
MDS_SENDTYPE_BCAST (attached hack ) is resulting osafimmnd Sync
MESSAGE:1005094 OUT OF ORDER , because of Immsv is sending some multicast
message size is more than fragmented size 65480, when we have large imm Objects
(1000K Imm) to sync.
Because of that most of the MDS_SENDTYPE_BCAST message are getting send as
multicast and some messages are getting send as unicast with frag_and_send
option, that is why we are hitting OUT OF ORDER at receiver.
To fix the issue, I adjusted the IMM constants to send broadcast message below
fragmented size 65480 and tested multicast feature with below two option (
please see #654 for more details about these options ) with 1000K Imm Objects
(1 million) and observed considerable performance improvement.
The test case I have chosen is multiple nodes simultaneously joining to an
existing
Active controller (all immnds should be in `IMM_NODE_LOADING` state at a
given point of time), this is one of the case where Immsv do BCAST to newly
joined immnd (S) to syn objects.
1) With using `BSRsock` socket for both unicast & multicast and Introduce
a new socket for multicast.
2) Introduce a new socket for multicast and use existing `BSRsock` sockets
unicast.
Following are the statistics:
1) With previous broadcast (implemented as multiple unicasts ) :
*************************************************************************************
Jun 23 20:28:12 SLES-SLOT2 opensafd: Starting OpenSAF Services
Jun 23 20:29:37 SLES-SLOT2 opensafd: OpenSAF(4.5.M0 - ) services successfully
started =====> 85 seconds
Jun 23 14:58:31 SLES-SLOT3 opensafd: Starting OpenSAF Services
Jun 23 14:59:55 SLES-SLOT3 opensafd: OpenSAF(4.5.M0 - ) services successfully
started =====> 84 seconds
Jun 23 14:53:28 SLES-SLOT4 opensafd: Starting OpenSAF Services
Jun 23 14:54:52 SLES-SLOT4 opensafd: OpenSAF(4.5.M0 - ) services successfully
started =====> 84 seconds
*************************************************************************************
2) With new multicast TIPC_ADDR_MCAST :
*************************************************************************************
Jun 23 19:32:55 SLES-SLOT2 opensafd: Starting OpenSAF Services
Jun 23 19:33:42 SLES-SLOT2 opensafd: OpenSAF(4.5.M0 - ) services successfully
started =====> 47 seconds
Jun 23 14:03:14 SLES-SLOT3 opensafd: Starting OpenSAF Services
Jun 23 14:04:00 SLES-SLOT3 opensafd: OpenSAF(4.5.M0 - ) services successfully
started =====> 46 seconds
Jun 23 13:58:11 SLES-SLOT4 opensafd: Starting OpenSAF Services
Jun 23 13:58:57 SLES-SLOT4 opensafd: OpenSAF(4.5.M0 - ) services successfully
started =====> 46 seconds
*************************************************************************************
---
** [tickets:#851] MDS: use TIPC multicast**
**Status:** assigned
**Milestone:** 4.5.FC
**Created:** Mon Apr 14, 2014 09:31 AM UTC by Hans Feldt
**Last Updated:** Fri Jun 20, 2014 06:52 AM UTC
**Owner:** A V Mahesh (AVM)
MDS broadcast is implemented as multiple unicasts. This does not scale and adds
unnecessary load to the sending process.
MDS broadcast should use TIPC multicast.
Attached is a hack that does this. It can be used to measure performance
improvements.
---
Sent from sourceforge.net because [email protected] is
subscribed to https://sourceforge.net/p/opensaf/tickets/
To unsubscribe from further messages, a project admin can change settings at
https://sourceforge.net/p/opensaf/admin/tickets/options. Or, if this is a
mailing list, you can unsubscribe from the mailing list.
------------------------------------------------------------------------------
HPCC Systems Open Source Big Data Platform from LexisNexis Risk Solutions
Find What Matters Most in Your Big Data with HPCC Systems
Open Source. Fast. Scalable. Simple. Ideal for Dirty Data.
Leverages Graph Analysis for Fast Processing & Easy Data Exploration
http://p.sf.net/sfu/hpccsystems
_______________________________________________
Opensaf-tickets mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/opensaf-tickets
