Typical discussion on alignment…
Me: “How many SAN IO Calls would you need to make to get File 1 for the Misaligned storage?”
Customer: 5
Me: “What about aligned?”
Customer: 2
Me: “What about File 2?”
Customer: 7
Me: “What about aligned?”
Customer: 2
Me: “How many tiny files, and fragments do you have on your fileservers?”
Customer: “Millions…”
Me: “Wouldn’t you like to save 2-4x or more IO calls and cache?”
Customer: “We love you, have our babies!”
Me: “P.S. Quit using fileserver VMs! Use a share off your NetApp! Give me the money you would have spent on MS licenses and CALs, I’ll make better use of it!”
OMFG BBQ SAUCE!
What impact can this have on the SAN? It’s like watching the school nerd run a Presidential Fitness Challenge 1 mile. It’s not pretty. In fact, it can bring it to it’s knees crying and panting. Run Forrest Run!
Dr. Spengler would say “Try to imagine all life as you know it stopping instantaneously and every molecule in your body exploding at the speed of light.”
You think I’m kidding, well I’m not.
If you max out CPU, if you max out NVRAM, your SAN is going to shit the bed. (If you never have seen a toddler shit the bed, they tend to use it like crayons and draw with it all over the walls and windows and crib. Ick. YOU will have to clean that mess up, because your poor administration foresight and planning would be the cause of a big mess! )
I have seen it. I have had to be the clean up crew in many customer environments. Especially due to P2V Windows, and wizard click click click linux VMs
Other notes:
- Misalignment is like beating a ginger up for their lunch money.
- VMFS3 tried to align to 32KB boundary
- VMFS3 to VMFS5 upgrade kept the 32KB boundary.
- VMFS5 tries to align at the 1MB boundary, like Windows 2k8 R2
- Linux (fdisk) on anything older than RHEL6/Cent6 muxxes crap up with a sector 63 start because of the good ol days of a single spinning disk with LBA.
- If you or a former SAN admin were the click click type, and you wizard’d your LUN creation, go back and make sure they are VMWare as the type (or the proper RDM type). I see ‘solaris’ as lun types all the freakin time since it was a certain default entry in a dropdown of a certain SAN manufacturer and people always over looked it!
- If you P2V a vm, you are probably misaligned.
I won’t go over how to detect misalignment. It’s 1am, and it’s easily google’d. And I may already have a blog post about it.
Required Reading:
- NetApp Tech Report 3747 “Best Practices for File System Alignment in a Virtualized Environment” for more information – www.netapp.com/us/library/technical-reports/tr-3747.html
- Misalignment of luns KB: https://kb.netapp.com/support/index?page=content&id=2013676
- “Partition Alignment” section in Performance Best Practices for VMware vSphere 5 http://www.vmware.com/pdf/Perf_Best_Practices_vSphere5.0.pdf
- “ESX Storage Considerations” section in Performance Best Practices for VMware vSphere 4.1 http://www.vmware.com/resources/techresources/10161
[asa]0321799933[/asa]
[asa]1479112569[/asa]
Comments are closed.