From: Conner McCleod (conner.mccleod@gmail.com)
Date: Fri May 11 2007 - 10:05:16 EDT
Solaris 10
IPMP three modes:
Probe-based IPMP with active-standby setup (IPMP-P-AS) AKA
active-passive .
Probe-based IPMP with active-active setup (IPMP-P-AA)
Link-based IPMP with active-standby setup (IPMP-L-AS)
What are the performance differences between IPMP-P-AS and IPMP-P-AA?
Do some applications fail with an IPMP-P-AA (two IPs plumb)?
Is there a way to setup the Link based IPMP to an active-active mode?
What is the difference between Link- based IPMP and Probe based IPMP?
How does Link-based IPMP compare to the other IPMP modes?
Which of these IPMP setups offers the most bandwidth and stability and
application compatibility?
Besides (nicstat 5) what other tests or tools would you use to test these
setups for Network performance?
I am working on several environments with mission critical setups. In the
past I have used Probe-based IPMP with Active-Standby setups and they worked
fine. However, most of my servers were network busy but not network
intensive. My new setup is network intensive with over 30 zones servicing
thousands of clients and each zone doing a different workload (one zone
Weblogic, another zone oracle, another zone web servers, another zone
Websphere etc etc). By far these new setups have the most network
requirements as I am combining the loads from 40 machines into 6 domains.
Which of the IPMP setups above offers the best solution for this kind of
setup?
I am told that some applications get configured with the active-active
setup. I am not able to reproduced nor have I seen the problem. Do you
know applications that will not work with an Active-Active setup?
Does Link-based IPMP works at layer 2 (link) to the switch? If that is the
case, is it possible that a link-based IPMP setup will have link 2 network
level outage and fail to switch... such as the network down but the port
still live? The Probe-based IPMP seems to have more checks build in (layer
2 and layer 3) as the actual test traffic of the probe IP must pass through
all the network layers.
I have read the posting on IPMP and some are very good. Seem like other
people have preferences for active-standby. I also read the Blueprint:
http://www.sun.com/blueprints/1102/806-7230.pdf However, I did not find
articles that compare the performance or discuss both the Link based and
Probe based IPMP. Though the Blueprint does suggest plumbing all NICs
(active-active).
Ay suggesting or helps as it relates to performance and the differences
between the setups are really appreciated. Here are the test setups that I
am using:
** Probe based IPMP (Active Active):
>>>>>>>>>> bge0 gets one IP and hme0 get another IP (two different IPs).
** Probe based IPMP (Active Standby):
>>>>>>>>>> bge0 and hme0 get the same IP.
**** Link Based IPMP:
>>>>>>>>>> bge0 and hme0 get the same IP.
The only difference on Probe based IPMP active-active' setup versus
active-standby's setup is the world "deprecated" on the second configuration
file ei (hostname.hme0). Here are the details of the TEST setups and the
configurations:
########################################
** Probe based IPMP (Active Active):
########################################
# cat hostname.bge0
168.109.98.60 netmask + broadcast + group PROD_ipmp_p2_0 up
addif 168.109.98.69 netmask + broadcast + deprecated -failover up
# cat hostname.hme0
168.109.98.68 -failover netmask + broadcast + group PROD_ipmp_p2_0
# netstat -rn
Routing Table: IPv4
Destination Gateway Flags Ref Use Interface
-------------------- -------------------- ----- ----- ------ ---------
168.109.0.0 168.109.98.68 U 1 7 hme0
168.109.0.0 168.109.98.60 U 1 1 bge0
168.109.0.0 168.109.98.60 U 1 0 bge0:1
224.0.0.0 168.109.98.60 U 1 0 bge0
default 168.109.98.2 UG 1 16
127.0.0.1 127.0.0.1 UH 4 1935 lo0
>>>>>>>>>> bge0 gets one IP and hme0 get another IP (two different IPs).
########################################
** Probe based IPMP (Active Standby):
########################################
# cat hostname.bge0
168.109.98.60 netmask + broadcast + group PROD_ipmp_p2_0 up
addif 168.109.98.69 netmask + broadcast + deprecated -failover up
# cat hostname.hme0
168.109.98.68 deprecated -failover netmask + broadcast + group
PROD_ipmp_p2_0
+++ notice the word deprecated on the above line.
# netstat -rn
Routing Table: IPv4
Destination Gateway Flags Ref Use Interface
-------------------- -------------------- ----- ----- ------ ---------
168.109.0.0 168.109.98.60 U 1 6 bge0
168.109.0.0 168.109.98.60 U 1 0 bge0:1
168.109.0.0 168.109.98.60 U 1 1 hme0
224.0.0.0 168.109.98.60 U 1 0 bge0
default 168.109.98.2 UG 1 5
127.0.0.1 127.0.0.1 UH 4 86 lo0
>>>>>>>>>> bge0 and hme0 get the same IP.
########################################
**** Link Based IPMP:
########################################
# cat hostname.bge0
test9860 netmask + broadcast + group PROD_imp1 up
# cat hostname.hme0
group PROD_imp1 up
# netstat -rn
Routing Table: IPv4
Destination Gateway Flags Ref Use Interface
-------------------- -------------------- ----- ----- ------ ---------
168.109.0.0 168.109.98.60 U 1 3 bge0
224.0.0.0 168.109.98.60 U 1 0 bge0
default 168.109.98.2 UG 1 5
127.0.0.1 127.0.0.1 UH 1 14 lo0
>>>>>>>>>> bge0 and hme0 get the same IP.
Conner
_______________________________________________
sunmanagers mailing list
sunmanagers@sunmanagers.org
http://www.sunmanagers.org/mailman/listinfo/sunmanagers
This archive was generated by hypermail 2.1.7 : Wed Apr 09 2008 - 23:41:57 EDT