Hiya,
I would liketo share with you some problems I've found in my environment.
I have tow clusters named as GOLD and SILVER.
GOLD cluster is compound by 3 ESX 3.5 and a EMC SAN.
SILVER cluster is compund by 5 ESX 3.5 and a BULL SAN.
A VCS call CONSOLE manage GOLD and SILVER clusters and it's attached to EMC and BULL SANs.
On BULL SAN we use a LUN for templates. We manage the same templates for both clusters. When I try to deploy a template to GOLD cluster, the process become very, very slow. I suppose that this deployment is done through network (a Gbps connection) between CONSOLE (VCS) and a GOLD ESX server... is it true?
How do you implement templates?
Maybe having a LUN on EMC SAN and scpying templates every time you modify it?
Can be templates on VCS and do deploy SAN to SAN?
Any comment would be great...
Thanks
If you want to keep the deploy time small, then you'll have to keep multiple copies of the same template. This isn't very efficient as now you can have the same changes not reflected in both templates. Since the SAN's are not replicating, to get an image from one to the other will require some sort of network copy. Whether that be from ESX when you deploy across SANs, or you scp, is up to you. Personally, I keep my templates in one place, and let the deploy task do the network copy for me. If I have to deploy multiple vm's, then I'll deploy one from template, and then clone the new vm into as many as I need. Depending on the size of your template, this process should usually be fairly quick. For this reason, an NFS datastore is also common to store templates, but with multiple SANs, unless you copy the template into all of them, network copy will be involved.
-KjB
If you want to keep the deploy time small, then you'll have to keep multiple copies of the same template. This isn't very efficient as now you can have the same changes not reflected in both templates. Since the SAN's are not replicating, to get an image from one to the other will require some sort of network copy. Whether that be from ESX when you deploy across SANs, or you scp, is up to you. Personally, I keep my templates in one place, and let the deploy task do the network copy for me. If I have to deploy multiple vm's, then I'll deploy one from template, and then clone the new vm into as many as I need. Depending on the size of your template, this process should usually be fairly quick. For this reason, an NFS datastore is also common to store templates, but with multiple SANs, unless you copy the template into all of them, network copy will be involved.
-KjB
buid a freenas box on each cluster , create an NFS datastore on it for templates , then set up rsync to keep them synchronised ?
Chris, doesn't this require that you create two templates? TemplateA on NasA connected seen by the hosts in ClusterA and then TemplateB on NasB connected seen by the hosts in ClusterB? Paying attention that when you deploy a template to clusterA you need to take TemplateA and not TemplateB (although they are the same because of the low level sync).
I was discussing a scenario like this with a colleague for very dispersed template deployments on clusters across the world...... where you let the storage replicate the templates.. but yet you have to artificially create a template infrastructure that allows for manually deploy the closest template to the cluster.
In a scenario like the one discussed in the thread isn't enough to create a single SHARED LUN across the two clusters (it could be a NAS share or it could be a LUN from either one of the two storage systems) where you have the templates which is then "local" to both clusters?
Massimo.
hmm I'm thinking of build ISO files... the same template can't be mounted in 2 places at once.
Unless.... you used a scheduled task to clone a template to a local ( to the cluster) NFS store ?
I am not sure I follow you.
I have depicted two different scenarios. The first is for a "localized" solution where the hosts of two different clusters would share a shared LUN on top of which you create a single template. You create the template once at the vCenter level and you can deploy it to both clusterA or clusterB. This should be technically possible.
The second scenario is for a "geographically dispersed" solution where one cluster is located in India and the other one in Italy (managed by a single vCenter). In this case I was imagining that you could create two templates one on a LUN of the storage in India and one on a LUN of the storage in Italy and setup LUN replication so that when you modify the "master" template in India the "slave" template in Italy gets updated automatically. Notice that master/slave is not a VCenter concept (where the two templates would be two independent entities) but rather a function of how you setup LUN replication. So the templates would be "logically" independent but "practically" identical in lock-steps. There would be another discussion in what you actually want to replicate (i.e. VMDK files) and not replicate (i.e. VMX files). It's the responsibility of the global administrator then to use the template that is closer to the cluster they want to deploy it to.
For the purpose of this thread I think the first scenario would be a better fit (if supported by the VI3 architecture, which I think it's the case).
Massimo.
I think the problem that the OP has is that he has 2 clusters with storage on seperate SAN's ( and I'm guessing no way of presenting a LUN from 1 SAN to the other ? )
Ah .. if that is the case than the scenario #2 is the only option.
Massimo.