[Last Call] Learn how to a build a cloud-first strategyRegister Now

x
  • Status: Solved
  • Priority: Medium
  • Security: Public
  • Views: 1111
  • Last Modified:

Change the blocksize of a VMFS filesystem

Hi,

We have 2 LUN that have been presented to an ESX 3.5 server. Both LUNS were formated as VMFS with the default 1MB blocksize for a VMFS filesystem.

We have now to place a big VMDK file in one of the Datastore (blank with no information named DS1) that goes beyond what the 1MB blocksize limit can handle.

Could someone walk us through the steps we have to follow in order to change the blocksize for this datastore? The Datastore does not contain any data and is a LUN that it has been carved on the SAN.

Thank you.
0
llarava
Asked:
llarava
  • 10
  • 9
1 Solution
 
ryder0707Commented:
You can unpresent the LUNs from storage, delete & recreate the LUNs, represent them then do a hba scan on the esx, add both LUNs as datastore and format with different block size, eg. 8MB

Or just remove the datastores, add both LUNs as datastore but this time reformat with 8MB
0
 
llaravaAuthor Commented:
Sorry I am not really familiar with this so could you please describe how do we do this:

"You can unpresent the LUNs from storage, delete & recreate the LUNs, represent them then do a hba scan on the esx, add both LUNs as datastore and format with different block size, eg. 8MB"

We just want to need this with one DataStore.





0
 
ryder0707Commented:
If you dont have access to storage its fine, just use step 2 above "Or just remove the datastores, add both LUNs as datastore but this time reformat with 8MB" both will get same result

If you want them as a single datastore, you have 2 options
1. delete both LUNs and recreate a single LUN with bigger space
2. create one datastore but use the extent feature(you can keep adding separate LUN to increase datastore space), this is like having one big logical datastore

which one you prefer? for option 1, you need to access your storage device to recreate & represent the LUN
0
Transaction-level recovery for Oracle database

Veeam Explore for Oracle delivers low RTOs and RPOs with agentless transaction log backup and transaction-level recovery of Oracle databases. You can restore the database to a precise point in time, even to a specific transaction.

 
llaravaAuthor Commented:
No I don't want to add them as a single datastore.

We have 2 datastores and each one in a different LUN. DS1 and DS2, we woud like to change the bit size for DS2. DS1 should remain the same since it's being used and have VM's.

Also I don't have access to the SAN so I have to work my way out with what I have...I hope that gives you more information. Summarizing the DS2 DataStore has to be reformated with a different blocksize so we can upload larger vmdk files.
 
0
 
ryder0707Commented:
OK go to Configration tab, click storage, ensure you are at the "Datastore view" not "Devices view" (2 small buttons at the top)
Look for DS2 & right click it and select delete, take note all data will be lost in that DS2
Once "recent tasks" confirmed deletion has completed & DS2 has dissappeared
Click "Add Storage" at the top right to add new datastore, select Disk/LUN and click next follow on screen instruction to specify DS name and choose 8MB as the blocksize
0
 
BijuMenonCommented:
You can login to ESX Service Console using Putty and use command line like one below to create a 2MB Block.
vmkfstools -C vmfs2 -b 2m -n 32 vmhba1:3:0:1

Command Line explained in the below article
http://pubs.vmware.com/vi301/server_config/wwhelp/wwhimpl/common/html/wwhelp.htm?context=server_config&file=sc_appb.21.5.html

1. 1MB block size = 256GB max file size
2. 2MB block size = 512GB max file size
3. 4MB block size = 1024GB max file size
4. 8MB block size = 2048GB max file size.
0
 
ryder0707Commented:
You might want to describe to him which vmhba to format otherwise he will loose all his data and by the way it is vmfs3 not vmfs2
vmfs2 is considered obsolete and in fact removing the module can actually improve hba scanning
The recommended way to reformat is via the VI client as the partition will be alligned
vmfs partition formatted in service console is not alligned and VMs can suffer performance issues
This is vmware best practice, if you still need to format in service console, you need to know how to allign the partition manually
0
 
llaravaAuthor Commented:
ryder0707:


2.jpg
1.jpg
0
 
llaravaAuthor Commented:
ryder0707:

From the images I posted above. What should be entered as the SAN Identifier.
You mentioned the name of the DataStore that was deleted? SO DS2?

Just want to clarify before taking any actions.

Thanks.
0
 
ryder0707Commented:
Yes remove but ensure you select the correct one and also check no data inside
You can use the datastore browser to verify

Click "Add Storage" at the top right to add new datastore, select Disk/LUN and click next follow on screen instruction to specify DS name and choose 8MB as the blocksize

You dont have to worry about SAN identifier, you will see all available LUNs in the list, select the LUN you want to add and click Next, you can give any name you want, just something human readable
0
 
ryder0707Commented:
Fyi, I normally use the following naming convention for datastore
StorageModel-DSfunction-RAID--TieringGroup-LUNid
for eg. EVA8100-WEBSQLDB-1-3-10
0
 
llaravaAuthor Commented:
Deliting the DataStore from the ESX sever will not delete the LUN on the SAN. Just double checking.

My understanding is that you will delete the DataStore from the ESX server and then you will select the LUN from the SAN that will be hosting the DataStore finally giving the format.
0
 
ryder0707Commented:
Nope but data will be lost, after deleting it, you can see it in the available LUN to add when you click "add storage under disk/LUN"

"My understanding is that you will delete the DataStore from the ESX server and then you will select the LUN from the SAN that will be hosting the DataStore finally giving the format. "
Not entirely correct, you dont choose the SAN to host the datastore, the datastore will be created on the host where you click "add storage"
For example the same LUN id is presented to multiple esx hosts(you already added it as DS), if you delete that datastore on any esx host, all hosts will loose that same DS
0
 
llaravaAuthor Commented:
Ok. So I have 3 ESX servers (ESX1,2 and 3) and they all see the same LUNS and have the access to the same DataStores. DS1 and DS2 (this is the one that should be deleted and reformated).

So If a delete DS2 from ESX1 will 2 and 3 see the change I've done? What are the steps that have to be done with the other servers if any?

By the way, I got the following from the SAN guy but I don't really understand how to match it up with ESX can you help?
3.jpg
0
 
ryder0707Commented:
Seems like those are the LUNs that are presented to vmware host
Trust me you dont have to worry about that unless you want to do troubleshooting at service console
If you are really sure DS2 is the LUN you want to remove & reformat with different block size, just do it as your 1st screenshot(pls verify no data inside before doing this)
next, go to each esx configuration tab > storage adapter > rescan all hba
verify DS2 has gone
now go to any esx host(ESX1,2 or 3)then click "add storage" to readd the same LUN as datastore
It will appear automatically on all hosts after you are done formatting it
0
 
llaravaAuthor Commented:
Yep. It worked like a champ...Thanks Ryder

In our case we are doing a P2V of a 300GB physical D: drive so we have seleted 8MB block size = 2048GB max file size. I guess that the biggest my VMDK will be able to become is up to 2TB (which is the max size for a Datastore).

Some questions about this:

1.) What do we loose resizing the block size to something bigger than the 1MB default one.
2.) If we ever over the 2TB limit size what can we create another VMFS DataStore and merge it with this one?
3.) Also we could migrate the physical server D: with all the DATA to a RAW LUN in the SAN an then present that to the ESX server. What we are doing is P2V the D: drive instead and bringing the VMDK to the VMFS DataStore. For us is easy to do the P2V D: than the copy D: to a RAW LUN, etc... which solution do you think is better. Wouldn't make more sense to use VMFS (VMware) than a RAW LUN?

Thanks.  
 
0
 
ryder0707Commented:
I guess that the biggest my VMDK will be able to become is up to 2TB (which is the max size for a Datastore).
Nope, max size for vmfs datastore is actually 64TB only the file within will max at 2TB. Refer to http://www.vmware.com/pdf/vsphere4/r40/vsp_40_config_max.pdf

1.) What do we loose resizing the block size to something bigger than the 1MB default one.
Nothing important unless you are using thin provisioned disk. Thin disk on 8MB will grow by 8MB increment, ultimately reducing metadata changes which mean better performance, no sure how big is the performance but it wont be that great i think
Dont worry, vmfls block size will not effect guest I/O(from what i know so far)

2.) If we ever over the 2TB limit size what can we create another VMFS DataStore and merge it with this one?
Yes possible by using extent but it is not recommended as failure of a single LUN will cause you to loose the entire datastore and it's also not possible to remove a single LUN after you've added it to extent
In vsphere4, you can actually grow the datastore, request LUN expansion from your storage admin, esx4 will detect the extra free space and you are able to grow to the max available with a few click in the vSphere client

3.) Also we could migrate the physical server D: with all the DATA to a RAW LUN in the SAN an then present that to the ESX server. What we are doing is P2V the D: drive instead and bringing the VMDK to the VMFS DataStore. For us is easy to do the P2V D: than the copy D: to a RAW LUN, etc... which solution do you think is better. Wouldn't make more sense to use VMFS (VMware) than a RAW LUN?
If you ask for my opinion, yes i think its easier to manage vmdk and according to vmware whitepaper, both vmdk & rdm produce similar I/O throughput, you might be interested with pvscsi instead
http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=1010398
http://www.vmwareinfo.com/2009/06/whats-deal-with-new-pvscsi-drivers.html
0
 
llaravaAuthor Commented:
unfortunately we are not yet in VSphere...how hard is to upgrade? Is the upgrade free?
0
 
ryder0707Commented:
First of all, licensing scheme has changed, you can login to your vmware account to retrive the new license key to upgrade to vSphere based on your current eligibility
Upgrade process is not that hard, basically upgrade the VCS, all hosts(easily done using host update utility), vmware tools on each VM, upgrade virtual hardware to ver 7 etc
but its a long process to explain here
Best is to refer to the official guide as it gives sample upgrade scenarios based on your current environment http://www.vmware.com/pdf/vsphere4/r40_u1/vsp_40_u1_upgrade_guide.pdf
0
 
llaravaAuthor Commented:
Excellent. Thank you.
0

Featured Post

Configuration Guide and Best Practices

Read the guide to learn how to orchestrate Data ONTAP, create application-consistent backups and enable fast recovery from NetApp storage snapshots. Version 9.5 also contains performance and scalability enhancements to meet the needs of the largest enterprise environments.

  • 10
  • 9
Tackle projects and never again get stuck behind a technical roadblock.
Join Now