Bonding <-> MPIO & Number of virtual IP-Adresses

[matthias23]

Hi there,

I got some questions regarding the Howto "V7 active-active iSCSI failover".

In the tutorial MPIO is used but on the storages there are two nics each bonded. Is this done because there are only two nics on each storage client to access the storages? My storage clients (Citrix Xen) have 4 nics each for the storageaccess so I could refrain from bonding and use 4 MPIO-paths? Right? Am I missing something?


Furthermore I'm not sure if I do understand why there are 4 virtual IP-adresses used in the howto. Of cause we need a own subnet for every MPIO-path, but why aren't 2 virtual IP-adresses (subnets) sufficient? Obviously there are 2 (different) paths for each iSCSI-Target, but why?

So, in my setup with 4 MPIO-paths, 2 logical volumes on two iSCSI-targets. Do I need 8 virtual IP-adresses (in 8 different subnets) for the two storages?

Best regards!
Matthias

[To-M]

We recomend the bonding for best practices but you dont have too have the bonding for the VIPs and you can use the 4 MPIO-paths.

The 4 virtual IP-adresses are the combination of the Local and Remote Nodes IPs.
With your setup you can have 4 virtual IP-adresses or more if you wish to have more then one stream and gain path failover with MPIO.

[matthias23]

Thanks.

But I don't get why you use 4 ip-adresses in 4 different subnets in your active-active example. XenServer 6.1 doesn't seem to support 2 IP-Adresses per interface by default.

Why dont use 4 adresses in 2 different subnets for example, e.g. 192.168.21.99 for target0 & bond0 192.168.21.100 for target1 & bond0, 192.168.31.99 for taget0 & bond1, 192.168.31.100 for target1 & bond1?

[To-M]

By not applying our recomendation with the 4 different subnets it will cause serious routing issues and in case of moving resources access to iscsi storage can be lost. Though you can do this but I wouldnt recomend it.