UNSOLVED
Need advice in setting up GEO MS cluster with SQL2008 connected to CX4-960
Hi All
We are in a testing phase of setting up GEO MS cluster 2008 running on DB SQL 2008 attached to an Array cx4-960. This is what we are trying to achieve but not sure if this setup will impact my inter-DC dark fiber link of 1GB which is already running with 100TB+ LUNs with Async mirror daily.
1) We have a CX4-960 at primary site with 2 nodes Active/passive mscluster2008 running with sql2008. We are adding 3rd to node to the DR site which is like 100 Kms far & we have 2 x 1 GB link running between primary to DR for mirrorview async replication. We have 2 MDS9506 switches running each site. ISL is configured using IVR & the link is 1 gb per switch between the 2 datacenter. Now the database on the original cluster at primary site is of 2.5 Tb with a 1gb quorom lun. Our windows administrators want us to zone the DR node 3 with primary CX4-960 which I can easily achieve by zoning the primary storage ports with DR 3 server node using the IVR zone & they want us to assign all the luns assigned to the orignal cluster in primary site to the 3rd node. (This I can easily achieve by adding the 3rd node in to the storage group of original cluster). Now my query is that at the movement we are just adding the cluster so definitely their wont be much of an activity on my 2 x 1GB link. but I am trying to figure out what would be the scenario if incase one or both the nodes at primary site goes down & we happen to fail the cluster resource to the 3rd node which is at DR site, would this support?. Would my 2 x 1GB link support the IOs which the 3rd node would generate if it becomes active. Any advice or any other way we could achieve this?
regards
Firoz
+447919381259
Storagesavvy
474 Posts
338
1
Posted March 8th, 2012 11:00
To answer your other question of whether there are other ways to achieve this... Yes, there are several different options.
First ask yourself what is the likelihood that both primary site nodes go down and the storage array itself is still available? Yes it's possible but assuming you put your two nodes in different racks, with redundancy power, and connect those hosts to redundant network and fabric switches, it is unlikely they would both go down unless something more drastic occurred in the datacenter.
You could put a second CX4 in the second site and connect the 3rd node directly to it. Then using MirrorView/S and Cluster Enabler (a plug-in for Microsoft Cluster Services) the mirror is automatically failed over and back when the cluster resources are moved back and forth. This eliminates the WAN latency and bandwidth concern from application performance after failover, and also provides complete redundancy of the storage and servers.
This is just one option, there are other options that may or may not be best for your environment and what you are trying to accomplish.
Storagesavvy
474 Posts
338
1
Posted March 8th, 2012 11:00
Assuming you have IVR and Zoning set up correctly for the 3rd Node so that it can see the LUNs the simple answer is yes, it will work. Regarding the performance over the 1Gbe links, that all depends..
There are two factors affecting performance -- Latency and Bandwidth.
Latency: Assuming your 100km dark-fibre link has about 2ms of round trip latency, then your 3rd node will have 2ms of additional response time for IO's beyond what nodes 1 and 2 experience. This will affect the over IOPS achieveable for that node. Higher latency reduces IOPS in general.
Band This translates to about 200MB/sec. Whether this is enough depends on the total MB/sec bandwidth your primary nodes require under normal circumstances, and how much bandwidth is available after you account for other unrelated traffic sharing the connection. Let's say that your inter-DC links are carrying 100MB/sec of traffic right now, then you only have about 100MB/sec left before you hit the limit. (In reality, driving WAN links to 100% is difficult to achieve without really good switch/dwdm/wan acceleration equipment.) You will need to check the current performance for the primary nodes and see how much bandwidth they are currently driving to the array.
If you don't have enough bandwidth, the net result is that the 3rd node will perform slowly which will drive up latency. The latency of the WAN link itself, plus any latency caused by over committment of the link is added to the inherent array/disk response time. As overall response time increases, there may be a limit to what your application can deal with. SQL and Exchange for example require response times around 20ms or less before problems start occurring.
While I'm not answering your question directly with a yes or no, hopefully this will help you answer it by running some calculations based on your environment.
Richard Anderson
firozg
41 Posts
338
0
Posted March 8th, 2012 01:00
Hello Techies...
No reply to my query yet?
regards
firoz