Design NTNX site


Badge +4
Hello,

We are in the process of moving our VMWare site to Nutanix. With 3 locations i would like some design tips to setup the NTNX sites/cluster(s). I am very new to nutanix.

Situation: 3 fysical sites, 2 sites with 4 NTNX hosts and 1 site with 3 NTNX hosts. Every site dedicated fiber connection. The 2 "4 hosts sites" are for production, the 3 hosts site will be otap and/or extra hosts for production.

How to design the NTNX cluster(s) with minimal risk of loosing data and the most redundancy.

If someone needs more info please let me know

grtz

11 replies

Userlevel 4
Badge +21
Hello,

It would help to know about your workloads/VMs:
- What's your actual design (How many VMs ? How big are they/how much VM data do you have on each site?, existing ESX/storage specs, replication already in place? etc.)
- Do you want/need to replicate your VMs from one site to another?
- If so, do you want to replicate all of your VMs between your 2 "4 nodes" sites? between each "4 nodes" site and the "3 nodes" site? Otherwise? How frequently do you want to replicate your VMs? Do you want to retain 1 or more previous versions of you VMs?
- Do you already own the Nutanix nodes/block? if so, what models do you have?

We can help you on your design, but we need basic information to do so!

Sylvain.
Badge +4
Hello Shuguet,

Thank you for the quick response.

In reply to your questions:

About 100 vm's with around 3TB data. All is now in one site with 5 hosts (esx 4.2). We do own the Nutanix blocks (NX3050) and ESXi 5.5 is installed on NOS 3.5.

Both production sites need to hold the same vm's and the same data all the time if possible. If one fysical site goes down the other site needs to take over with no loss of data or vm's if possible.

The otap site is for testing but should also handle production if needed (perhaps one host OTAP and 2 hosts production).

In scematic we want to do the following:

Location A - Production

Location B - Production

Location C - Production / Testing
Userlevel 4
Badge +21
There is still many options here, depending on your network configuration.Do you have the same subnets on each site?If not, you may need to change the IP adresses (perhaps also update your DNS records in this case)?One possible scenario here would be to setup replication between the two "4 nodes" sites (1h RPO is the current lowest option).You would then create 2 containers (1 on each site) setup replication, put all your VMs on one site (to ensure consistency across all your VMs, I would not suggest splitting you VMs in 2 halves and running half on one site and the other half on the other site).Depending on your needs and space availability on the third site, you can also setup replication from the two "4 nodes" sites to this one, or only for some VMs.Sylvain.
Badge +4
Hello,

Subnets is another question, the desing is not yet made so we are very flexible.

Does you're scenario not mean that all the load is on one fysical site and the other fysical site does not have any load or does vmware handle the load (if it is one big cluster of production)?

And how about the production data? If one fysical site goes down is than production data consistant on another fysical site or is the data max 1hour behind?
Userlevel 4
Badge +21
Yes, in the scenario I described, all your workload will be on one physical site, and none on the other.I proposed this in order to avoid possible data inconsistencies as much as possible (but no, in this scenario, you don't have "one big cluster", you have 2 clusters, one on each site).For the moment (it may change in the future) Nutanix only offers 1H RPO, so that means that whichever of your site goes down may have data from the other site as much as 1H behind.Take 2 applications depending on each other for example. If you put the first application on site A and the other application on site B, if site B goes down, the local copy of the second application may be as far behind as 1H, thus you will have break the timeline between application 1 and application 2.For this reason and this reason only did I suggest this Active/Passive scenario.If this is not a problem for you, you can have half of your VMs on each site (that may be a good scenario if you already have application-level replication for example).With the currently available Nutanix functionalities, there is nothing you can use to have always synchronous data on both site, as there is no "data locality" or "RF locality" in the NDFS code right now.It may (and will as I hear it) be available in the future tho.For now, you will need third-parties tools like Zerto to provide "nearly synchronous" replication, if this is really what you want.One thing to keep in mind though is that with synchronous replication, if you delete something on site A, it's also instantly gone on site B, so you will need some sort of backup/asynchronous replication to prevent this kind of problem.Sylvain.
Badge +4
ok, is there an other scenario that we can use, because this is not what we want.

I was told that every nutanix host writes data with an RF of minimal 2 and that it could stretch fysical sites so that a metro cluster is not needed and data is allways available to the vm even when one fysical site is down.
Userlevel 4
Badge +21
You may have to contact your local Nutanix SE to discuss this, because as far as I know there is currently no public information about site locality.(Note: I'm not working for Nutanix, so maybe there is and I didn't see it).Sylvain.
Userlevel 4
Badge +21
According to this blog post: http://nutanix.blogspot.fr/2014/05/40-features-fault-domain-awareness.htmlYou may be able to use this feature to have the design you want.I'm not sure how it will be implemented though, because the 4.0 is still not publicly available.Sylvain.
Badge +4
ok

Tnx for discussing this with me.

I will contact a Nutanix SE and talk about this with him/her.
Badge +4
tnx,

This means that wat we want should be possible with NOS 4.0.

NOS 4.0 should allready be available states the press release(by the end of april), but they may have hit some bugs or performance problems because it still is not available 😞
Userlevel 4
Badge +21
No problem, it is my pleasure.

Do not hesitate if you need any more help on Nutanix.

Sylvain.

Reply