I ran some iometer tests on both netapp vmdk and a cx380 vmdk and the netapp outperforms the cx380 vmdk by far. I am not sure why this is.
I do not use powerpath on out esxi servers so which there is 4 paths active, only one shows as i/o active.
on hte netapp it is set to round robin and both paths have active i/o.
but why would the netapp outperform the emc by so much?
the netapp is a 3210 on an aggregate with 14 disks dp.
the cx380 has 2 luns on 2 rgs
1 is a sata R5, 7 disks
2 is a FC R5 5 disks
both are using FCP
The number of IOPS you can do are directly proportional to the number of spindles you throw at it. The NetApp should bury the CX380 simply because the CX380 doesn't have enough spindles. This is doubly true if the NetApp aggregate is using SAS disks.
Assume that a 7.2K SATA disk is capable of 100 IOPS and a 15K SAS drive is capable of 180 IOPS.
If your 3210 has an aggregate made up of SATA disks, it's going to top out at 12*100, 1,200 IOPS. If they're SAS, the high end is 2,160 IOPS.
Your CX380 RG1 will top out at 600 and RG2 will top out at 720.
It's not a fair fight of the controllers since the controllers are capable of far more I/O but you've constrained the testing by the number of available spindles.
Hello.
Note: This discussion was moved from the VMware ESXi 5 community to the VMware vSphere™ Storage community.
What NetApp are you using? What storage protocols? How many disks are backing the datastores? Which firmwares are in use on each storage system? The answers to any of these might help explain, as there are many reasons that could lead to this.
Good Luck!
the netapp is a 3210 on an aggregate with 14 disks dp.
the cx380 has 2 luns on 2 rgs
1 is a sata R5, 7 disks
2 is a FC R5 5 disks
both are using FCP
Well, the single active path would definitly impact that.
If you configure the pathing properly so you have equivlant paths, and then also put a similar number of disks behind the LUNs, you'll see performance get closer.
What you've asked is: "I have a Corolla with a V6 and a Maxima with a inline 4 - why is the Corolla so much faster".
the netapp is a 3210 on an aggregate with 14 disks dp.
the cx380 has 2 luns on 2 rgs
1 is a sata R5, 7 disks
2 is a FC R5 5 disks
both are using FCP
The number of IOPS you can do are directly proportional to the number of spindles you throw at it. The NetApp should bury the CX380 simply because the CX380 doesn't have enough spindles. This is doubly true if the NetApp aggregate is using SAS disks.
Assume that a 7.2K SATA disk is capable of 100 IOPS and a 15K SAS drive is capable of 180 IOPS.
If your 3210 has an aggregate made up of SATA disks, it's going to top out at 12*100, 1,200 IOPS. If they're SAS, the high end is 2,160 IOPS.
Your CX380 RG1 will top out at 600 and RG2 will top out at 720.
It's not a fair fight of the controllers since the controllers are capable of far more I/O but you've constrained the testing by the number of available spindles.
I am using mru (vmware) to the cx380. Is this the best option? I do not have powerpath
thank you for the clarification. that helps a lot
I am using mru (vmware) to the cx380. Is this the best option? I do not have powerpath
MRU is rarely the best option although I'm not a VMware/EMC expert. Try Round Robin instead.
MRU says to use the same path over and over again until it failus. Round Robin spreads the I/O across your available paths.
Round robin is NOT supported on the CX3.
I am doing a similar test on a vmdk that sits on a cx340 6TB RG that has 14Disks (SATA, 4GBps) R5
ran iometer with 16K 75% Read
and the performance is still bad. compared to netapp.
total io/s is about 500
total mbs is 8.6
avg io res time is 58
max i/o response time 2266
terrible
the 6 tb rg is carved up into 10 other luns but the other luns are very minimally used. pretty muych empty luns.
on the netapp
I get
over 20000io/s
320mb/s
1.5ms av resp time
max io resp time 115
there is such a huge difference