2013-05-12 02:53 PM
All our LUNs are presented to ESX 4.1 and are VMFS. Our Netapp Performance Advisor says we have about 40%-50% misaligned reads and some writes on all our LUNs. It also adivses us that every single LUN is violating LUN partial reads and says we should make sure the host file system has block sizes in mutiples of 4k. However, our NetApp VSC advises us that every single VM in our cluster is "actually alligned". We created all our LUNs using OnCommand System Manager, type is always set to "VMware". All our VMs are Windows Server 2008 so should be aligned out of the gate. All our LUNs are formated VMFS with a block size of 8MB (cleanly divides by 4k).
Where else can we be going wrong here? According to everything I can find we are aligned except according to Performance Advisor...any help would be greatly appreciated.
2013-05-12 03:26 PM
Also should note, I just checked a LUN that has a SQL database on it. Performance Advisor shows a real good 1% misaligned reads and 0% misaligned writes. So I am guessing that the OS alignment is just to ensure that misalignment is not 100% of the bat. So what is considered a "normal" misaligned read/write percentage on an aligned Server 2008 VM? The SQL data LUN that is doing so well was created the same as all other LUNs. The only difference is the data on it. Despite the SQL LUN reporting a low 1% misaligned reads it still is reported in Performance Advisor as being misaligned. Is there always going to be some misaligned reads/writes? And if so why would performance advisor scream about only 1% misaligned reads? Seems that no matter what almost every LUN will be reported as misaligned even though its not...
2014-07-23 10:03 AM
Sorry to necro this thread, but I'm seeing the same thing here.
I'd say our VMs are 97% Windows 2008 or 2012, 1% Win2k3, and 2% other.
On one LUN, I am seeing 75% misaligned reads & 50% misaligned writes. Another LUN (Citrix Xenapp servers on Win2k8) has 93% misaligned writes and all over the place with reads, between 10% and 80%. Another LUN is lower, around 20% misaligned reads & writes. The Exchange 2013 LUNs are beautiful with almost zero misaligned WAFL ops.
Netapp Balance reports 10 VM's out of alignment. We have 250+ VM's, so 10 is not a big percentage. However, in vSphere we don't have the option to align the VM's that Balance says aren't aligned.
Other info that might matter:
Our volumes were rebuilt after upgrading to ESXi 5.x to get them upgraded to VMFS5.
The LUNs on one controller of an HR pair seem to be less aligned than its partner. The controller with more misaligned traffic also has quite a bit more overall traffic in general.
We're using FAS 3270 with a bunch of 600GB 15k drives, some SATA 3TB + 4TB, and some 1.2TB 10k 2.5" drives. I believe we have two 1TB flash-cache modules. Data ONTAP 8.1.3.
VMware is all v5.1.