HPE MegaRaid configuraion on ProLiant DL380 Gen10 Plus

[Matej2292]

Hello,

 

Recently we recived new ProLiant DL380 Gen10 Plus servers. In ILO we configured MegaRaid and logical drive this way on both servers:

ProLiant Gen10 Plus - MegaRAID configuration (002).png (udrop.com)

These Gen10 Plus were built and configured as an VMware ESXi hosts (Windows is not a running OS on these servers ). 

These hosts have FC cable - 10 Gbit/s connected directly between them and on vSwitch we configured it as a vmotion for both hosts. 

The problem is that migration via vMotion is too slow considering the fact that it should be 10Gbit/s. It's taking too long to migrate even small machines between these two ESXi hosts. Seems like the migration run through 1Gbit/s instead of 10 Gbit/s.

VMkernel adapter and also physical adapters are configured correctly and we also tried many troubleshooting steps and test but nothing help-ed. So we suppose that bottleneck is cause by this MegaRaid which is a new topic for us.

Is there some way how to configured MegaRaid and logical drive so migration through 10Gbit/s will be functional ?

Does anyone came through the similar problems?

Thank you in advance.

 

[a_p_]

Do you migrate the VMs in powered off, or powered on state?
Prior to ESXi 8.0, cold migration (powered off) used the NFC protocol with a peak performance of ~1.3 Gbps.

see e.g. https://core.vmware.com/resource/vsphere-vmotion-unified-data-transport

André

[Matej2292]

We try to migrate Vms in powered on mode. It works well in other environments where is no MegaRaid configuration and vMotion uses 10Gbit/s connection.

Version of our ESXi hosts is VMware ESXi, 7.0.3, 20036589

 

Thank you.

[a_p_]

Assuming that the vMotion is configured properly, i.e. using the 10Gbps network, and the RAID controller has write-back cache, another step to troubleshooting to check what esxtop (see e.g. https://www.yellow-bricks.com/esxtop/) reports.

André

[Matej2292]

Hello Andre,

Thank you so much for your advices.

We already tried to check esxtop command on affected ESXi hosts.

Here is a screenshot:

https://www.udrop.com/7F1y

As you can see vMotion is configured on vmnic1 and there are no dropped packets.

Could you please give me some advice what else we can check in esxtop reports?

We have a suspucion that bottle neck is caused by MegaRaid configuration, but we are not sure how to configure it correctly...

[DanielKrieger]

Hi, what Raid Level and Discs do you use?

Is your Raid Controller in the HCL List?

You can check with ESXTop your Disks (Press D for Disk Mode).

 

Cheers Daniel

[TomasSnajdar]

Hello Daniel,

we are using official configuration of ProLiant, which is directly predefined by HPE so I believe that all components are in HCL List, same for VMware (MR416i-a is supported by VMware vSphere 7).

Regarding RAID configuration, as this is smaller system, we use RAID1 with 2x 2,4TB disk.

 

[Kinnison]

Hi,

Let me ask a question, say RAID 1 but without specifying the disk drive technology is not of particular use, please could you specify this detail.
Maybe I'm wrong but I hope we are no be talking about mechanical 12G SAS drives, personally SSD drives with a capacity of 2.4 TB I don't remember seeing any.

Regards,
Ferdinando

[Matej2292]

Hello Ferdinando,

These 2.4 TB drives are SAS HDD Media type.

Thank you for your help.

[DanielKrieger]

Hi your Disk performance is really bad with 2 SAS HDD Disks in Raid 1. I think you will run in a Bottleneck. With this raid, there is no chance to use the 10GB/s Network.

Can you migrate VMs in power off state?

Have you check your Disks with esxtop while migrating?

https://kb.vmware.com/s/article/1008205

Here you can calculate your Disk Performance

https://wintelguy.com/raidperf.pl

 

[Kinnison]

Hi ,

An excerpt from the official documentation:

"In vSphere 7.0 Update 1 or earlier, vMotion saturates 1 GbE and 10 GbE physical NICs with a single vMotion VMkernel NIC. Starting with vSphere 7.0 Update 2, vMotion saturates high speed links such as 25 GbE, 40 GbE and 100 GbE with a single vMotion VMkernel NIC. If you do not have dedicated uplinks for vMotion, you can use Network I/O Control to limit vMotion bandwidth use".

Which is absolutely true but on the condition that the resources and performance of the starting and destination systems, especially as regards storage allow for it. Now, simplifying (the topic is complex and there are many other factors at play), with a RAID 1 array with two drives like yours and even with the help of your "controller cache" you will never saturate a 10GbE link, rather you don't even come close to it.

Let's make it simple, create a new virtual machine with a "Tick provision eager zeroed" disk of say 100/150 Gigabytes, then monitor the real-time performance of your storage system (esxtop option d) calculate the average and you will get a more realistic idea about the performance of your storage and thus what will be reasonable to expect when you perform a "compute and storage vMotion operation".

Regards,
Ferdinando