Unsolved
This post is more than 5 years old
8 Posts
0
141063
May 25th, 2010 14:00
Making sense of the IOPS numbers
I've been running vFogLight Pro for almost half a year now, and I have yet to be able to correlate the IOPS to what I see in ESXTOP. I could really use some help understanding this, since so far I find that I can't rely on vFogLight (and that obviously isn't a good thing).
So I have a VM which is heavy on IOPS. When I'm in ESXTOP, the VM will regularly spike between several hundred to several thousand CMDS/s (IOPS). We would see it hover around 4k - 5k, for example. However, in the last 4 hours, vFogLight's Commands graph shows spikes approaching 240k and some smaller spikes over 80k. The Storage tab is showing 8k Commands Issued in the Physical Disk table. Additionally, the Monitor tab and the Physical Drives box in th Storage tab both show the same number, but it's different than the Physical Disk table. They show 82 reads and 20 writes. When I click on that box, it gives me that graph which goes up to 240k. When I drill into the Physical Disk table, I get the same box as in the Physical Drives box.
So I'm confused. Where am I supposed to look to get an accurate representation of the current IOPS for VM? My guess is that it's in the Physical Drives box and the Physical Drives table shows IOPS for the entire LUN across all the hosts. But is that roll-up data or current? And if I'm correct about the Physical Drives box being the current value for the VM, how can I explain the 240k IOPS spikes? Is that really legitimate? If so, then I guess I'm impressed by my storage device!
![]()
A related question: say I have the 8k of IOPS for the LUN. Is there a report or view which gives me a breakdown of what VMs are causing those IOPS? I know I can get a page which shows me the VMs on the LUN, but I want one that shows me that VMa is taking 20%, VMb is taking 5%, and VMc is taking 75%. This will help me know who's doing the damage without having to drill into each VM and do the math.
Thanks, all! I'm trying to find more ways to realize returns in our very heavy investment in vFogLight.
Brian
So I have a VM which is heavy on IOPS. When I'm in ESXTOP, the VM will regularly spike between several hundred to several thousand CMDS/s (IOPS). We would see it hover around 4k - 5k, for example. However, in the last 4 hours, vFogLight's Commands graph shows spikes approaching 240k and some smaller spikes over 80k. The Storage tab is showing 8k Commands Issued in the Physical Disk table. Additionally, the Monitor tab and the Physical Drives box in th Storage tab both show the same number, but it's different than the Physical Disk table. They show 82 reads and 20 writes. When I click on that box, it gives me that graph which goes up to 240k. When I drill into the Physical Disk table, I get the same box as in the Physical Drives box.
So I'm confused. Where am I supposed to look to get an accurate representation of the current IOPS for VM? My guess is that it's in the Physical Drives box and the Physical Drives table shows IOPS for the entire LUN across all the hosts. But is that roll-up data or current? And if I'm correct about the Physical Drives box being the current value for the VM, how can I explain the 240k IOPS spikes? Is that really legitimate? If so, then I guess I'm impressed by my storage device!
A related question: say I have the 8k of IOPS for the LUN. Is there a report or view which gives me a breakdown of what VMs are causing those IOPS? I know I can get a page which shows me the VMs on the LUN, but I want one that shows me that VMa is taking 20%, VMb is taking 5%, and VMc is taking 75%. This will help me know who's doing the damage without having to drill into each VM and do the math.
Thanks, all! I'm trying to find more ways to realize returns in our very heavy investment in vFogLight.
Brian
No Events found!
brian.laws
8 Posts
0
May 25th, 2010 14:00
bmalone11
33 Posts
0
April 24th, 2012 13:00
I'm seeing similar unexpectedly high figures from vfoglight for the iops per vm and per datastore which make me suspect the data is either calculated incorrectly. This is one of the few statistics in vFoglight which cannot easily be gathered from vCenter, so it would be a shame if the data is wrong....does anyone out there use vFoglight IOPs data and trust its accuracy?
bmalone11
33 Posts
0
April 26th, 2012 15:00
Any clarity would be much appreciated on this.
vFoglight is telling me my average cluster IOPS over the past hour is around 7000. The SAN which the cluster is connected to is reporting about 500-600 IOPS over the same period. The latter figure is more believable considering the disk configuration we are using.
Where is vFoglight getting this data from?
DELL-Thomas B
171 Posts
0
April 27th, 2012 13:00
Both are correct actually... The VMware SDK for Reads/Writes is the total # of each for a given interval, we use the Real-time API which uses a 20 second interval. So to truly understand this as a rate vs. a count you need to divide the number by 20. So when you see 7000 commands (reads/writes/total) and divide that by 20, which would get you about 350. The rest on the SAN side is probably for other things attached.
bmalone11
33 Posts
0
May 15th, 2012 09:00
Hi Thomas,
that's helpful to know. What's misleading is that the view says Total IOPs (average) which implies io per second, rather than per 20 second! By your calculation, in the data below, the average total IOPS is actually 528 rather than 10,560.
DELL-Thomas B
171 Posts
0
May 15th, 2012 18:00
Indeed, I'm updating the cartridge and changing the calculations to reflect this more accurately.