Hello
Firstly, thank you for developing this tool it has been very informative for us so far!
I am trying to understand the 4KB alignment ratio graph. I assume this is something to do with making sure the guest VM's partition or the ESXI host and what ever your storage subsystem isnt mis-alligned causing split IO? I am not really sure how to interpret this graph. Does this look healthly or point to potential issues? As this is the first time we run the tool I have no idea of what is normal.
The graph in question comes from a Windows 2008 R2 VM running on ESX 6.0.
The disk in the guest is Partition Starting Offset 1,048,576 bytes
Running the following command "fsutil fsinfo ntfsinfo f:" gives me the below output.
Number Sectors : | 0x000000001f3fe7ff |
Total Clusters : | 0x0000000003e7fcff |
Free Clusters : | 0x00000000026abb60 |
Total Reserved : | 0x0000000000000000 |
Bytes Per Sector : | 512 |
Bytes Per Physical Sector : | <Not Supported> |
Bytes Per Cluster : | 4096 |
Bytes Per FileRecord Segment | : 1024 |
Clusters Per FileRecord Segment : 0
Mft Valid Data Length : | 0x0000000000500000 |
Mft Start Lcn : | 0x00000000000c0000 |
Mft2 Start Lcn : | 0x0000000000000002 |
Mft Zone Start : | 0x0000000000fe7ba0 |
Mft Zone End : | 0x0000000000ff42c0 |
Also could someone clarify on Throughput as we have seen this spike as 25000 (MBps which would be 25GB of data in a short window? ) for read & write and on the same time frame IOPs spike uptp 40/50k for read & write for very short periods of time. It sounds like its the am
You can go throught the following link https://blogs.vmware.com/vsphere/2012/05/troubleshooting-storage-performance-in-vsphere-part-1-the-b... this would help in understanding poor storage performance is generally the result of high I/O latency
Regarding 4K alignment ratio:
As you interpreted, this graphs shows whether the I/Os are aligned to 4KB addresses or not. If the total value (red line) is always at 100%, that's the ideal situation with all I/Os aligned to 4K. Any value below 100% results in a read-modify-write or I/O splitting in the case of 512e(emulated) or 4Kn(native) drives. This implies a potential scope of improvement in workload application. But, in the case of 512n drives, this does not necessarily imply an issue. For example, a 512byte I/O which is not aligned to 4K, but aligned to a sector boundary doesn't result in any IO amplification. In some cases, this can also be due to potential issue and worth looking at it if there are significant performance regressions.
(You may ignore the vertical blue/green lines as the ratio can become zero in intervals when there is zero read/write)
Regarding the throughput spikes:
As per the graphs, there are ~50k ops of 512KB size resulting in 50GB/s and this seems unlikely. Would you please share more details on the workload that was run on the VM, drive type etc?