That's not correct - one IP Address will utilize the LACP trunk itself - as you are assigning the IP Address to the virtual LACP device not to individual NIC port. Say for example -

you create one LACP trunk with 4 ports - on the data mover, using cge0, cge1, cge2 & cge3 - the LACP device is given a virtual name say for example trk0. Now while configuring IP Address, you have NO option to assign the IP address to the individual cge ports as they are part of the logical LACP device. You can only assign the IP Address on the LACP device i.e. trk0 in this case.

By this way, multiple clients or users can use one IP address but the IP Address is built on a virtual device of aggregate 4 GiBps bandwidth.

In my previous post, I mentioned that if you want, you can configure multiple IP Addresses on the virtual LACP device i.e trk0 here - if needed - and each IP address (interface) will effectively use the virtual LACP trunk.

However, keep in mind that although LACP is providing aggregate bandwidth and high availability - one given TCP session can not utilize 4 gig bandwidth as each session will be based on wire speed only i.e. Gigabit ethernet speed.

Am I able to provide some clarification? or creating more confusion?
Please revert back,
Thanks,
Sandip

If you have multiple NICs configured in a LACP - no need to use multiple IP Addresses for the LACP - you can simply use one IP Address for the LACP device and mount the NFS export using that IP Address.

However, if you want to have multiple IP Addresses - you can do so. But they are going to be configured on the same LACP device (in fact you will not be able to configure IP Addresses to individual NIC (cge ports) once they are configured with LACP). With that configuration, you can use any of these IP Addresses for the NFS mount as long as they are in the same subnet or VLAN. All of the IP addresses will eventually utilize the LACP trunk.

Hope this provides some clarifications - if not, please fel free to revert back.
Thanks,
Sandip

Thanks Sandip! To clarify if I only utilize one IP address for the target mountpoint then regardless of the number of devices in my LACP data will always only flow through one device at a time correct? The goal is for higher throughput as well as availability. Thanks!

I am glad to be able to provide some clarification on the IP Address configuration and use of LACP channel (trunk) on Celerra.

On the second topic - any port-channeling (LACP or Etherchannel) provides aggregate bandwidth combining all the available NICs - but can not provide a cumulative bandwidth for a given TCP Session - this is Network standard, nothing to do with Celerra. Also, I am not sure, whether any TCP session really needs 4 Gig pipe for its own - so LACP can be very well be utilized for your VMware datastore.

There is one more feature on Celerra - called Fail Safe Network - which is a high available solution - it configures one physical or logical device as primary and another physical or logical device as standby - which can provide you switch level redundancy as well.

Lastly - Celerra does support 10G port - it comes in specific NAS model only. I suggest you to get in touch with EMC TC or other contacts.

There are few technical documents available on using VMware data store on Celerra using NFS or iSCSI. You may consult your EMC Technical Consultant or other EMC contacts for more details and guidance.
Thanks,
Sandip

Sandip one of your last points was what I was referring to, I realize you do not assign IP addresses to the individual interfaces that are members of the LACP configuration:

"... one given TCP session cannot utilize 4 gig bandwidth as each session will be based on wire speed only..."

So your saying that a VMware NFS datastore sitting on a Celerra with only 1 IP address referencing the datastore on the Celerra can utilize more than one of the paths in the LACP configuration, but an individual session will never utilize more than one of the paths?

So really you never have 4 gig of bandwidth available for a single session. Is there a way to have 4 gig of bandwidth available for a single TCP session? Can the session be broken down transmitted across all available paths then reassembled? So my question is can we get 4 gig of throughput for a VMware datastore residing on a Celerra with LACP configured, or would that require 10G nics on both the Celerra and switch side, which I am pretty sure Celerra does not have yet?

Thanks again!

as Sandip said - this is a fundamental thing because of the implementation of LACP and NFS / TCP/IP

while there are different load-balancing option in LACP one NFS mount will always use the same IP,MAC and port number - so for one mount (data store) you will only use one physical link

I think this was done by the switch implementers to make it easier on their ASICs and the buffer space.
If they would put one NFS packet fragmented on more than one link then buffering and reassembly if one link is a bit slower gets pretty challenging

What you can do is to create four data stores and mount each one from a different IP - then they togehter can use the links fully

Thanks Rainer!! This is exactly what I was looking for, I saw the post on iSCSI a couple months back and was waiting for this. Thanks again!!

your're very welcome
Rainer