Two things off the top of my head after skimming the given details:

1. iSCSI will work better without the bond. It already uses multipath, so all you need is to separate the portal IPs/subnets and provide separate IPs/subnets to the iSCSI dedicated NICs, as is the recommended way here: https://access.redhat.com/solutions/131153 and also be sure to follow this: https://access.redhat.com/documentation/en-US/Red_Hat_Enterprise_Virtualization/3.6/html/Administration_Guide/sect-Preparing_and_Adding_Block_Storage.html#Configuring_iSCSI_Multipathing

2. You haven't mentioned anything about jumbo frames, are you using those? If not, it is a very good idea to start.

And 3: since this is RHEV, you might get much more help from the official support than from this list.

Hope this helps

Dan

On Fri, Jun 24, 2016 at 9:12 PM, Colin Coe <colin.coe@gmail.com> wrote:

Hi all

We run four RHEV datacenters, two PROD, one DEV and one TEST/Training. They are all working OK but I'd like a definitive answer on how I should be configuring the networking side as I'm pretty sure we're getting sub-optimal networking performance.

All datacenters are housed in HP C7000 Blade enclosures. The PROD datacenters use HP 4730 iSCSI SAN clusters, each datacenter has a cluster of two 4730s. These are configured RAID5 internally with NRAID1. The DEV and TEST datacenters are using P4500 iSCSI SANs and each datacenter has a cluster of three P4500s configured with RAID10 internally and NRAID5.

The HP C7000 each have two Flex10/10D interconnect modules configured in a redundant ring so that we can upgrade the interconnects without dropping network connectivity to the infrastructure. We use fat RHEL-H 7.2 hypervisors (HP BL460) and these are all configured with six network interfaces:
- eno1 and eno2 are bond0 which is the rhevm interface
- eno3 and eno4 are bond1 and all the VM VLANs are trunked over this bond using 802.1q
- eno5 and eno6 are bond2 and dedicated to iSCSI traffic

Is this the "correct" way to do this? If not, what should I be doing instead?

Thanks

CC

_______________________________________________
Users mailing list
Users@ovirt.org
http://lists.ovirt.org/mailman/listinfo/users