HPE StoreVirtual Storage / LeftHand
cancel
Showing results for 
Search instead for 
Did you mean: 

Long distance multisite config performance

bart verberne
Occasional Visitor

Long distance multisite config performance

Hello,

When designing a multisite config, I cannot find any info of the influence on total performance of a configuration when inter-site latency is substantial.

IE what is the performance of a 6-node multisite san (3 times P4500 SAS per site) when the distance is 50KM.

tnx

Bart

4 REPLIES
teledata
Respected Contributor

Re: Long distance multisite config performance

2 Points about Multi-Site SAN here:

1) Multi-site SAN configuration should not have any significant impact on performance.

2) A multi-site SAN REQUIRES a GbE connection with < 2-3ms average latency
http://www.tdonline.com
bart verberne
Occasional Visitor

Re: Long distance multisite config performance

Yes your statement is correct (we are planning to use 10Gbe between sites) but consider this:

as an example, and this is how I think it works : Imagine 2 nodes configured with network raid 10 with a network latency of 0,5 mS one-way, so 1 mS round-trip.

Server writes to node 1, who commits the write to cache and copies the write to node 2. when the write is committed to cache in node 2 it in turn replies an ok back to node 1, who then replies an I/O complete to the server.

I assume this dataflow is the case as it is standard practice in other arrays like an EVA.

So, this latency limits the maximum I/Os in this example to 1000 per second because of the 1mS round-trip latency.

Thats the reason I assume distance (latency) limits the performance of a multisite config, and I am looking for more specific information.
For instance I am not sure if this latency limits the overall performance of the cluster, or the performance of a node-pair.

is there a sizer or other info ?

Bart
teledata
Respected Contributor

Re: Long distance multisite config performance

I was just pondering this, and as I understand it, unless you are suffering a local node failure you should be accessing the local VIP and local storage nodes only, but I understand the concern about increased latency while the I/O is being replicated to the (N+1) nodes at the remote location.

I guess the real question is, DOES an I/O transaction wait for confirmation of the Network RAID copy before an ack is sent back to the client?

Technically speaking your entire IO transaction can be completed on the local site, and only a mirror copy of your datablocks is sent over the WAN (Assuming all links and nodes are up and healthy).

I'll ask around as well, but if you find out anything be sure to post it back here!

http://www.tdonline.com

Re: Long distance multisite config performance

Hello,

Any updates regarding this topic?
I would also like to know the answer.

Regards,
Eric