Stretching Failover Cluster (3 nodes) and P4300 (4 nodes) over 2 buildings

[Fred Blum]

Hi all,

I want to stretch our hardware over two buildings in order to have redundancy in case of disaster site failure. We have a 3 node W2008 R2 Failover Cluster with a 4 node Lefthand P4300 SAN. The distance between the buildings is around 150 meters apart.

Can someone point me into the direction of best practise documents for setting up such a configuration?

I have looked at the ProCurve 3500yl-48G-PWR with 10Gbe ports.  

TIA,

Fred

[oikjn]

Are you going to control the link between the two buildings?  If you can provide enough of a bandwidth trunk between the buildings, then I wouldn't even bother with setting up a multi-site cluster unless you get bandwidth limited.  The extra latency will be minor at that distance.  The only important thing to do is if you number your nodes 1,2,3,4...  you need to make sure your in your network raid stripe that the raid1 part splits the buildings which means you would keep the evens on one site and the odds on the other site (for a 4-node cluster).

I haven't tried, but you might be able to get away with setting up a multi-node cluster with three sites and keep the FOM+MS cluster nodes in the third site so they don't only use the nodes in your bilding and use both sides.

[Fred Blum]

Hi,

I am talking to providers that are offering dark fibre (private fibre channel). However that is quite a pricy solution. I have to talk to the council if it is allowed to connect the buildings by digging up the street and at what price.

What is the minimum bandwith needed to avoid setting up a multi site cluster?

TIA,

Fred

[David_Tocker]

At least 1gb if you are going to be running as a single cluster/MG.

If you can run fibre at a resonable cost it would be the best bet.

UBNT have 1.4gb airfibre which may do the same job and not require you to dig up the street, as long as you are willing to trust a wireless link (admittedly at 24ghz it is not the same kettle of fish as a normal wifi connection)

It basically works like a wireless patch lead and apparently introduces very-very minimal latency.

We are investigating this option at the moment.. a one off cost of about $2k usd is pretty attractive vs dark fibre and our experiences with UBNT have been very positive with their 802.11 gear.

If you were to go down this route i would recommend an online UPS at each end dedicated to the wireless link.

Hell, for that money you could run redundant links in active/active (lacp) very easily and get 2gb potentially.

[Jess1313]

Here is the P4000 HA/DR User Guide with best practices.

Hope it helps

http://bizsupport2.austin.hp.com/bc/docs/support/SupportManual/c01727773/c01727773.pdf

[Fred Blum]

We have applied for a permit to lay a dark fibre connection  between the buildings.

Can you elaborate on the pros and cons of "multi-node cluster with three sites and keep the FOM+MS cluster nodes in the third site so they don't only use the nodes in your bilding and use both sides."

TIA,

Fred

[oikjn]

the MS HP DSM provider is site aware and will only use the nodes at the backup site as a backup to reduce the amount of traffice across the link.  This is extreemely helpful when dealing with multi-site clusters spanning across state lines hundreds of miles apart where latency becomes an issue.  In your case, you have two sites that would have no real bandwidth limitation and no real latency limitation so ideally you would want to have your hosts talk to all nodes to make the most use of all nodes.  The catch is that unless you want to go with a double mirror on a four node cluster (NR10+2 which wastes a TON of space) you can maintain a single mirror (NR10) by making sure your striped date is in one building and the mirror data is in the other.  The multi-site cluster makes sure this is done in the background no matter how you order your cluster, but in the alternative you can achieve the same results by putting the odd cluster nodes in one building and the even in the other... you can test this by shutting off one node and you will notice that the data reduncancy dependency warnings will not show up on the other node at that building and only on the nodes of the other building (where the mirror data resides). 

Going multi-site make it easier to ensure that you don't mess up the cluster organization and accidentally lose your building resiliency.  For a 4-node cluster I chose to just manually order my nodes to ensure I got the resilliancy I wanted.  Just keep in mind that unless you have a true third location, if you ever lose the fiber link, the side with the FOM will control the cluster and the other side will go offline.  I would suggest you put a FOM in a 3rd location and connect it to the group using another medium... The latency and bandwidth requirements are minimal for the FOM I would setup a wireless link between the buildings as a backup and have the FOM routed over that network to both buildings and assuming you only have two buildings to work with and not a third, I would just put the FOM in a physically different location than the server room with the VSAs like a telco closet or even use a small quite server placed in your office area. 

[Prakash Singh_1]

Hi,

For P4000 SAN solutions querries you can also visit the HP Guided troubleshooting tree.

Below is the link for HPGT:

http://h20180.www2.hp.com/apps/Nav?h_pagetype=s-905&h_lang=en&h_cc=us&h_product=304617&h_audience=smb&h_client=s-h-e010-1&h_audiencerestrict=true&lang=en&cc=us