VMware Cloud Community
rwgorrel
Contributor
Contributor

SCSI Bus Sharing and RDMs... does it apply?

In this thread, it was suggested to me that the SCSI Bus Sharing setting of a virtual SCSI controller has no relevance to RDM's attached to that controller... that perhaps this setting only applys to attached vmdks/virtual disks. I was wondering what the vmware storage experts have to say about that?

I have some VM's setup as nodes of a Novell Cluster. The cluster resources are presented to the VM's as RDM's. I was told to change the SCSI Bus Sharing to from None to Physical in this setup, though I'm not exactly why. Of course, the major drawback of this is that once the SCSI controller is set to Physical Bus Sharing, the VM is no longer vmotionable. I would like to set this back to the None for vmotionability, by I'm unsure of what impact this has to Novell Clustering though some experience has shown me the nodes appear to operate just fine with None set as the SCSI Bus Sharing.

So does anyone have insight as to whether this setting has effect on RDM devices?

Thanks

-Rob

5 Replies
RParker
Immortal
Immortal

In this thread, it was suggested to me that the SCSI Bus Sharing setting of a virtual SCSI controller has no relevance to RDM's attached to that controller... that perhaps this setting only applys to attached vmdks/virtual disks. I was wondering what the vmware storage experts have to say about that?

That's incorrect. The ESX host where the VM lives is sharing with other VM's. If that's the case, then they are sharing the SCSI bus. RDM means they can gain access to the hardware at a different level than other VM's, but they most certainly share the bus.

Also change the bus to Virtual (unless there is a very compelling need to actually be physical.. and you already asked why, there is no good reason). Then you can vmotion them. Otherwise you are good.

0 Kudos
Texiwill
Leadership
Leadership

Hello,

For any shared disk cluster whether novel, RHEL, MSCS use the http://www.vmware.com/pdf/vi3_35/esx_3/vi3_35_25_u1_mscs.pdf as a guideline for setting up the virtual hardware. Pay close attention to the examples as the steps in the examples are very important and not necessarily within the text.


Best regards,

Edward L. Haletky

VMware Communities User Moderator

====

Author of the book 'VMWare ESX Server in the Enterprise: Planning and Securing Virtualization Servers', Copyright 2008 Pearson Education.

SearchVMware Blog: http://itknowledgeexchange.techtarget.com/virtualization-pro/

Blue Gears Blogs - http://www.itworld.com/ and http://www.networkworld.com/community/haletky

As well as the Virtualization Wiki at http://www.astroarch.com/wiki/index.php/Virtualization

--
Edward L. Haletky
vExpert XIV: 2009-2023,
VMTN Community Moderator
vSphere Upgrade Saga: https://www.astroarch.com/blogs
GitHub Repo: https://github.com/Texiwill
0 Kudos
rwgorrel
Contributor
Contributor

So if I am reading the MSCS document correctly, it advises setting the SCSI Bus Sharing to Physical... as I have done and was directed to do by the Novell documentation, though it still doesn't offer much of an explanation as why this is the case. Can someone explain it in more simple terms on why the SCSI Bus Sharing should be set to Physical when doing Multihost Clustering with a pass-through RDM's? does this mean you give up vmotionability for the nodes when you setting up this type of cluster? Because so far thats what I'm taking from this. I would like to understand the technicial implications of using None or Virtual in this configuration since it seems to work in at least some capacity.

-Rob

0 Kudos
rwgorrel
Contributor
Contributor

That's incorrect. The ESX host where the VM lives is sharing with other VM's. If that's the case, then they are sharing the SCSI bus. RDM means they can gain access to the hardware at a different level than other VM's, but they most certainly share the bus.

Well, normally, I have these virtual nodes (4 of them) scattered across 4 ESX servers, so it isn't that all the nodes live on the same ESX host sharing its bus. However, I would like to be able to vmotion each of these nodes to any of the other ESX hosts live and whether or not the ESX host is already hosting a different virtual node. All 4 of my ESX hosts are presented with the same access/paths to the SAN LUNs used by the Novell cluster.

-Rob

0 Kudos
Texiwill
Leadership
Leadership

Hello,

So if I am reading the MSCS document correctly, it advises setting the SCSI Bus Sharing to Physical... as I have done and was directed to do by the Novell documentation, though it still doesn't offer much of an explanation as why this is the case. Can someone explain it in more simple terms on why the SCSI Bus Sharing should be set to Physical when doing Multihost Clustering with a pass-through RDM's?

That is correct. The difference between None, Physical and Virtual is:

None implies no SCSI bus sharing, I.e. the RDM can not be used as part of a cluster

Virtual implies that the RDM will be shared by VMs on the same ESX host.

Physical implies that the RDM will be shared by VMs on different hosts

The issue is how the cluster will do locking. If its Virtual then its completely within one host and locking is done at the virtual scsi bus. If its physical, locking is done on the hardware with scsi-passthru locks. None, means the locks are handled entirely by ESX. Which is not what you really want for a cluster.

does this mean you give up vmotionability for the nodes when you setting up this type of cluster?

Yes.

>Because so far thats what I'm taking from this. I would like to understand the technicial implications of using None or Virtual in this configuration since it seems to work in at least some capacity.

Clusters and VMotion are mutually exclusive as you generally have your boot disks for the cluster nodes on local storage. This is part of one of the examples. While it does not need to be (changes in 3.5 I believe) it is still best to place them there.

IN essence, if you want the cluster to work properly you should use physical bus sharing for cluster nodes on different ESX hosts. You loose VMotion capability but you do not really need it if the nodes are on different hosts. Also, note no two nodes can be on the same hosts. You must maintain them on different hosts else the locking gets messed up.


Best regards,

Edward L. Haletky

VMware Communities User Moderator

====

Author of the book 'VMWare ESX Server in the Enterprise: Planning and Securing Virtualization Servers', Copyright 2008 Pearson Education.

SearchVMware Blog: http://itknowledgeexchange.techtarget.com/virtualization-pro/

Blue Gears Blogs - http://www.itworld.com/ and http://www.networkworld.com/community/haletky

As well as the Virtualization Wiki at http://www.astroarch.com/wiki/index.php/Virtualization

--
Edward L. Haletky
vExpert XIV: 2009-2023,
VMTN Community Moderator
vSphere Upgrade Saga: https://www.astroarch.com/blogs
GitHub Repo: https://github.com/Texiwill