DR-Site Design

[techaccesspak]

Dear Fellows,

We are making a Disaster Recovery site of our Primary Site (Data Center). I have questions that I am confused in:

1) Network Design

      We have many servers that must be connected to core. What is the best model to design a scalable data center. 

  a) The traditional 3 layered Model (Access/Distribution/Core Layer).

  b) The Top of the Rack Design Model. (1 TOR switch per rack and all the TORs are connected to Core)

2) What will be the IP Address Schema in DR-Site. Should it be same or different ? Currently, we are using private address schema (192.168.x.x/16) in our whole data center. This question is very important because of the Database replication that would be done from exisiting data center to DR-Site.

3) How should Existing Data Center be connected to DR-Site for database replication.

  1) Dark Fiber Direct Link. (It may not be possible)

  2) Connectivity through MPLS cloud.

I will be grateful if you could also share any helpful document/link regarding DR-Site data center.

I highly appreciate your time and help. Thanks.

[Philip D'Ath]

How  many servers is "many"?  10, 100, 1000?  How many racks are there?  3, 30, 300?

The applications you want to DR - can they do this at layer 3 (and manage having different IP addresses at each site)?  If so, then you can use different subnets at each site.  If not then you have to use a stretched layer 2 design.

Are you going to run the DR sites active/active or hot/standby?

It is impossible for us to tell you how much bandwidth is needed.  That will depend on all of the above, the amount of data to be replicated, and what you recovery time expectations are.

[techaccesspak]

Dear Philip,

Thank you very much for your quick response.

We have got like 20 applications with their databases. Some databases might require same subnet for replication of their databases to DR-Site. In this case, what can we do ? As mentioned before, we have two options:

1) Either we take a direct link (Dark Fiber). I am not sure if this can really happen as the distance between two sites is about 450 Km.

2) We get MPLS services from MPLS Provider. Either a layer 2 VPN (VPLS) or Layer 3 VPN.

Keeping in mind that we have Cisco 3800 ISRs on our Edge on two Sites.

Our Core is currently made of 6500, Our edge is 3800 with ASA/FWSM etc... In addition to this, we are using SAN, there are lots of applications that uses storage from multiple vendors.

What do you mean by a stretched layer 2 design. How can we acheive it in our case.

Thank you very much for your help.

[Philip D'Ath]

A stretched layer 2 design simply extends the VLANs from one site to another site.  So DR is very easy with this design.  You can simply restore a backed up machine from one site at the DR site and away it goes.

From a service provide you want to ask for a QinQ circuit.  This is a layer 2 Ethernet service that tunnels all of the VLANs from one site to another site.

You should also lay in redundant WAN and Internet tails into your DR site.

[techaccesspak]

Dear Philip,

Thank you for the explanation. Please correct my if I am wrong.

By QinQ, you mean VPLS / Ethernetover MPLS ?

Thanks.

[Philip D'Ath]

QinQ is a tunnelling technology that makes all the VLANs at one site appear at the other site - they are all fully connected.  QinQ looks like a dark fibre connection (except it only supports Ethernet frames).  Great for DR.

VPLS is a layer 3 (routed) technology, so I don't think it is suitable for what you have described.

You can also do tunnelling over MPLS, such as ATOM.

[Philip D'Ath]

You could also consider using Cisco LISP technology as well.  This article describes how it is done using dual data centres like what you want.  LISP can run over ordinary layer 3 routed connections, such as VPLS.

http://www.cisco.com/c/en/us/td/docs/solutions/Enterprise/Data_Center/DCI/5-0/LISPmobility/DCI_LISP_Host_Mobility/LISPmobile_5.html

But my first choice would be a QinQ link.