ESX VI3 support for Mount points on LUNs

[Pisapatis]

Can someone confirm that ESX supports mount points? For example, I have today 250GB LUN for VMFS. As we grow in number of VMs, I would like to expand VMFS file using mount points. Normally the solution would be to expand the existing LUN or extend the VMFS file over another LUN. What is supported in ESX OS? Is there a documentation?

Thanks in advance.

[bister]

Yes that is correct. Map another LUN to the ESX and just add another volume with the VIC: Host -> Configuration -> Storage -> Add Storage ...

Give it a name and you'll be able to put more VMs on it.

Regards,

Christian

[Pisapatis]

Will it be a separate VMFS or can it be part of the first VMFS (to utilize leftover)? If it is a separate VMFS, is it visible to another ESX server that is part of cluster or VMotion member? It seems to be more administration burdon if we add multiple VMFS file systems. Is there a way to extend one VMFS? Thanks.

[Gabrie1]

You can extend a volume over two LUNs. On VMFS2 you would lose the whole volume if one of the LUN would fail. But I thought I read somewhere that with VMFS3, you would only lose the data on the LUN that failed and the vmdk that would cross both LUNs (or would VMFS3 even have a protection against that?)?

Is it realy necessary to extend a volume? Would you have that much lost space? We use 500GB LUNs at this time and always calculate about 35GB of spare space for Snapshots and VM Swap.

Gabrie

[piacas]

Only use extents in an emergency for space. they are not a long term solution. Use extents to buy you some time while you provision a new LUN and then migrate VM's to new LUN.

[bister]

We use multiple LUNs at <400GB, all our ESX hosts see all LUNs, so Vmotion is possible between each host.

I see no administrative overhead in having one big LUN against multiple LUNs: I create it once and that's it.

Regards,

Christian

[Pisapatis]

What is the recommended size for VMFS to accomidate 20 VMs with Windows 2003 OS (4GB to 16GB RAM) that include swap files etc? What would be the VMswap size for each VM? Do we need to allocate additonal space for Consolidated backup?

Thanks.

[Jae_Ellers]

Since you didn't specify disk size of the vms I'll have to guess:

20 x 20 Gb virtual disk = 400 Gb

20 x 16 Gb swap = 320 Gb

\----

---

I'd give it 800 Gb lun for all together. Adjust swap calc to average swap size.

I'd also probably tend to whack this up into a couple 400 - 500 Gb luns just for flexibility/ease of migration & replication.

[Pisapatis]

Thanks for the recommendations. Is vSWAP file dynamic? I think it should be the part of VMFS that holds the VMs, right? What will happen if the vswap file is not enough? For example, I have 20 windows 2003 VMs (4GB-16GB) and I allocated only 300 GB for VMFS. When is vSWAP used by ESX kernel? I have 12 dual-core processors.

[Jae_Ellers]

I'm sure I'll get corrected if I'm wrong, but I believe the swap file now gets created at VM startup with the size of the allocated memory.

Cd to /vmfs/volumes/

ls -la \*/*.vswp and compare to allocated RAM.

Swap space is just for virtual memory for the vm.

[RParker]

that's correct. The swap is allocated at the time of creation.

If you add a swap later, the ESX host would need to be rebooted, and then additional vSwap space also get's created at that time. The vSwap reserves the space immediately, it's not like a Windows pagefile that gets dynamically allocated upon need, and hope available space is there when you need it.

[RParker]

The vSwap is additional RAM for VM.

If you have an 8 gig ESX host, and you have 10 VM's that all have 1g each allocated, then the ESX host would need 2gig of the 8 gig swap file. You can allocate more swap then you need, normally it's equal to the size of total Physical RAM.

So 8gig ESX host would have a 8 gig vSwap allocated. If the VM's run out of swap, it would create a condition where the VM would not start, because there is no RAM available.

[Rumple]

In VI3 the only swap you manually create is at install time and that is the Service Console Swap. You no longer create the additional swap within ESX like in 2.X anymore as each VM creates its VSWAP which is the same size as each VM RAM allocation.

If you have 10VM's at 1GB RAM each, then you are going to use 10GB space on your VMFS LUN for vswap (1GB in each VM directory)

[Jae_Ellers]

If you have an 8 gig ESX host, and you have 10 VM's

that all have 1g each allocated, then the ESX host

would need 2gig of the 8 gig swap file. You can

allocate more swap then you need, normally it's equal

to the size of total Physical RAM.

There's no monolithic vmwap structure with Esx 3.x. Each Vm has a swap file and from what I see it gets reserved as soon as the vm starts.

There is a console swap partition, but totally unrelated. In ESX 2 there's a swap file that gets created for the vms.

[Jae_Ellers]

Looking thru this I discoverd a windows vm with a 0 size vmsp. What could be the cause of this? It's up, running, and active.

[Pisapatis]

Though the physical RAM is limited, ESX allows to over-commit RAM requirements. For example, if the physical RAM = 32GB, and if three VMS are configured for 16GB, 16GB and 4GB, I think the effective swap file would be 36GB as against 32GB. Am I correct? If this is right, when all three VMs are active, I think there will be long disk activity that impacts performance. In this case, I think it is better to allocate vswap files on a seperate storage group (may be on RAID 10). Is it possible in ESX to seperate vSWAP and VMDK files?

[Rumple]

Currently I believe they have to be in the same directory for vmotion requirements,etc.

Also, I believe you can oversubscribe like 2.X as the memory subscription is similar I think.

[Jae_Ellers]

I think you have to edit the vmx config file directly to move virtual swap.

See http://www.vmware.com/community/thread.jspa?messageID=494967&#495040

Message was edited by:

Jae Ellers