SheGeek - it's not that simple. NFS is definitely the simplest option to present a datastore to an ESX host (several clicks less in the VMware GUI, no zoning/masking) - but you can't deploy MSCS/WSFC clusters without use of RDMs, no SRM support until later in 2009, and the performance stuff on the internet is only true for small-block IO. There are two TCP sessions per datastore, so scaling up (without 10GbE) means use of many NFS datastores and each will only fill one 1GbE connection. Thanks
BTW - that's coming from an unabashed pro-NFS supporter. In my opinion, having VMFS and NFS datastores (and small number of tactically used RDMs) for VI/vSphere clusters is the most flexible option. It also offers an easy way to get data onto and off your VMware environment, without using SCP tools.
It's a fundamentalist position to say "one way, always, regardless!" - call me a pragmatist!
It's better that if you got a unified storage which supports both FCP/iSCSI and NFS, so you can provision a NFS datastore for vmdks, and RDMs for these VM running cluster configurations. Running multiple VMs in a single LUN is not a good idea as there are I/O locks on the block side, but NFS don't have this issue.
Adawen - I tend to agree, and am happy that EMC offers exactly that capability. Your note about Locking/VMFS is unfortuately inaccurate and FUD (propagated with people with an NFS-only agenda as it suits them)
Details are on Virtual Geek
Each protocol has good/bad - choice and flexibility is good, along with technical clarity/accuracy.
Great job, very informative... I plan to do more research and see where in our environment we can take advantage of this offering.
pongraphan 1 year ago
Good job, easy to follow and understand! You also covered some great points for reviewing WMWARE studies.
LittleBrotherBlues 2 years ago 3
FCP is so not the best way to present storage to VMWare! You can get much better performance and manageability with NFS.
SheGeek1024 2 years ago
SheGeek - it's not that simple. NFS is definitely the simplest option to present a datastore to an ESX host (several clicks less in the VMware GUI, no zoning/masking) - but you can't deploy MSCS/WSFC clusters without use of RDMs, no SRM support until later in 2009, and the performance stuff on the internet is only true for small-block IO. There are two TCP sessions per datastore, so scaling up (without 10GbE) means use of many NFS datastores and each will only fill one 1GbE connection. Thanks
sakacc 2 years ago 4
BTW - that's coming from an unabashed pro-NFS supporter. In my opinion, having VMFS and NFS datastores (and small number of tactically used RDMs) for VI/vSphere clusters is the most flexible option. It also offers an easy way to get data onto and off your VMware environment, without using SCP tools.
It's a fundamentalist position to say "one way, always, regardless!" - call me a pragmatist!
sakacc 2 years ago 2
It's better that if you got a unified storage which supports both FCP/iSCSI and NFS, so you can provision a NFS datastore for vmdks, and RDMs for these VM running cluster configurations. Running multiple VMs in a single LUN is not a good idea as there are I/O locks on the block side, but NFS don't have this issue.
adawen 2 years ago
Adawen - I tend to agree, and am happy that EMC offers exactly that capability. Your note about Locking/VMFS is unfortuately inaccurate and FUD (propagated with people with an NFS-only agenda as it suits them)
Details are on Virtual Geek
Each protocol has good/bad - choice and flexibility is good, along with technical clarity/accuracy.
sakacc 2 years ago
Putting VMWare and shared storage together really solves some key pain points. Chad does a great job describing this environment.
pcaviness 3 years ago 2