Flex i/o module and iSCSI multipathing

Hi,

As per your setup, It will hardly show any difference in performance, Also note that in between SPA and SPB there is no redundancy among the Switches. As a workaround you may try to install another Switch 3 and share your ESXi hosts between Switch 1, Switch 2 and Switch 3.

Cheers,

Bashanta Phukon

I'm not sure why you say we would not see an increase in available iSCSI bandwidth. Can you elaborate?If I currently have a 1GB uplink from each SP to the switch and I want to bind and utilize all four ETH ports for that purpose, why would I not see a performance gain from such a change?

Hi jkirkbride

I was wondering if you can use LACP with iSCSI. Following EMC VNXe High Availability White Paper they used LACP only for NFS Setup and not iSCSI.

br

Patrick

Hello Patrick,

You can use LACP for iSCSI if you want, but iSCSI multipathing is better than LACP… (finer packets loadbalancing than LACP)

Regards, Benoît

Thanks for the reply. I guess what I am trying to figure out is if I can use LACP and iSCSI multipathing and see a performance increase. Right now, with multipathing, I get 2GB of bandwidth using two paths with a 1GB link each. If I use LACP along with a flex module to create two bound 4GB trunks, wouldn't that effectively increase to a total of 8GB of bandwidth with multipathing? That's what I am thinking. I'm looking for someone to tell me it will or won't work and why. Perhaps I'll need to go to EMC with this.

How many Discards do you have?

For me Discards occur when the Packet are recieved to late and therefore *discarded* because it cannot be used for the reconstruction of the whole data. So it is related to latency and not bandwith.
Have you tried to enable jumbo frames for the iSCSI Path to reduce the number of Packets which have to be sent?

Jkirkbride, i was wondering if your existing iSCSI Links are fully saturated. On the other side you are not increasing the connections from the server to the switch. In addition the VMware RoundRobin LoadBalancing switches the links it uses for traffic on a fixed time intervall. EMC recommends using PowerPath to have a Multipathing Software in place which provides more intelligent levels of LoadBalancings

As far as i understand LACP it gives you 8x1GB for Load Balancing and Rendundancy but not one big 8GB Pipe, what if i understand correctly,is what you want to have.

To increase the bandwith for one single link you need to upgrade to 10Gig Ethernet Infrastructure

Patrick,

Thanks for the additional info. One of the main indicators that I have seen in the environment we have  now is Transmit Discards on the switch ports leading to the interfaces on SPA and SPB. My thinking is that the SAN ports are being saturated which is why I am looking to increase bandwidth across multiple ports. I have also considered using more ports in the ESXi hosts but it didn't seem to be where the bottleneck was. I have thought about going to 10GB ethernet and we even priced it out but I need to make a really good case for it in order to justify the added expense. I wouldn't mind going this route and I could probably sell it if I have something conclusive that says that LACP and iSCSI multipathing won't do what I need.

Discards run in the 5,000-10,000 range per day, per interface. We are already using jumbo frames. From a latency perspective, disk latency from the hosts to the storage tends to be very low: averages are around 5ms for some datastores with most under 1ms.

Which Switch do you use?

Switches are HP 3500YL.

And btw, thanks for taking the time to respond.