Hello Tim,
> Nevertheless Nic's comment got me thinking about my current setup, so
> I thought I'd ask: is there any preference from a target perspective
> on whether you use multiple target IPs as I have, or just one single
> portal with an LACP port channel behind it?
The best practice you can find in my previous post and it scales. I
would stick with that setup.
If you use LACP you have to be aware of the following: ESXi 5.1 is the
first version that supports LACP and only with distributed vswitches and
only one LACP channel per ESX server. With distributed vSwitches you can
_not_ use iSCSI port binding with distributed vSwitches. When you use
LACP (Link aggregation control protocol is only about negotiating the
hash algorithm not about splitting up the load) there are two wide
spread load balancing methods:
- Source MAC hash (mac hash)
- Source and Destination TCP port and IP address hash (ip hash)
The problem with both the algorithms is that the same IP connection goes
over one link in order to avoid out of order transmissions (which would
slow it down even further. This is why all hash algorithms make sure
that the same stream (tcp connection) goes over the same link). And I
think you already know what that means: It means you limit the
bandwidth to the speed of one link per ESX. So now you can of course
create multiple portals, but that will only work if you use the ip hash.
Or you can configure the linux machine and create multiple virtual mac
address for each portal ip addresses which creates higher complexity.
Summary: Use ip port binding and multiple portals with PBR on the target
to get the maximum frontend throughput.
> Obviously I guess a lot of it will depend on the multipathing
> behaviour of the initiator, but wanted to ask if there were any
> gotchas from a target point of view.
Meassure it. Two commands I use very often are:
dstat - to the see the network throughput.
iostat -xm 2 - To see the backend I/O throuput I stole that from Nic.
Btw:
On the target I currently use 4x 1GBit links, two do lacp mac hash and
than I do the PBR with two portals over it. That limits the traffic to
two GBIT per ESX. If I wanted to have 4 GBIT frontend per ESX server, I
would go with 4 portal ip addresses and pbr, but currently I do not want
that. A better option is of course 10 GBIT, but that requires a higher
investment.
Cheers,
Thomas
--
To unsubscribe from this list: send the line "unsubscribe target-devel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
