topic Re: Gateway balancing in StoreVirtual Storage

Gateway balancing

M.Braak — Tue, 06 Dec 2011 08:55:12 GMT

Hi,

How does SAN/IQ exactly balance gateway connections? Based on what?

The reason i ask this is because i have 2 VSA's running on two ESXi servers. The VSA cluster has only one volume which holds a VMFS datastore and is two-way replicated.

Both ESXi server have a gateway connection with the same VSA.

When we look at performance counters of the VSA disk in vCenter i see all reading is done from the same VSA and writing is done to both (because of two-way replication)

Seems to me each ESXi node should use a different VSA for best performance? For now reading is done from only 50% of the disk spindles.

When i rebalance both connection stay to the same VSA.

Best Regards,

Marcel

Re: Gateway balancing

Dirk Trilsbeek — Tue, 06 Dec 2011 10:09:07 GMT

Hi,

the SAN/iQ-Software does load balancing by assigning volumes to different nodes in a cluster. If you only have one volume, there is no load balancing as all connections to that volume are handled by a single node.

Re: Gateway balancing

M.Braak — Tue, 06 Dec 2011 10:13:23 GMT

ah ok that explains it!

But then again i understand why gateway connections are going to one node but why are reads are done from one node?

For as i know the gateway server should also query the other vsa for data and then return to the client?

Now only 50% of the disk spindles are used for reading..

Re: Gateway balancing

Dirk Trilsbeek — Tue, 06 Dec 2011 11:14:40 GMT

i can only assume that this behavior is preferred for larger environments with several volumes. Data from the second node would have to travel by network to the node handling the volume, using some of the network bandwidth on both nodes. By reading everything from one node, there is more network bandwidth available over all nodes. Of course that won't help you as your second node only receives writes from the first node and never has to do any reads of its own. But assuming you had several volumes, this behavior should be faster.

Re: Gateway balancing

Bart_Heungens — Tue, 06 Dec 2011 11:56:36 GMT

Hi,

This is typical behavior of how ESX works with the VIP load balancing...

When you take a look at Windows servers, when installed with the HP DSM, there will be read and write actions on all nodes in the cluster...

For ESX, that DSM is (still) not there... Many rumours but no release dates yet...

To avoid to have many trafic on eventual ISL inter site links, you should work with sites... Then you can control the trafic to keep it on 1 site and spread eventually the load accross multiple nodes...

Kr,

Bart