ESXi Performance

[CTSTech]

In the Performance Tab, what are some graph chart counters to check for good/bad performance from an ESXi Server?

[vMario156]

Nearly all graphs are interessting and usefull

If you focus on a single ESXi host the CPU & RAM utilization is a mayor point.

It depends of course of the usecase: Are you doning some (performance) troubleshooting (what kind of problemes do you have?) or just interessting in some stats?

Regards,

Mario

[CTSTech]

vMario156 wrote:

It depends of course of the usecase: Are you doning some (performance) troubleshooting (what kind of problemes do you have?) or just interessting in some stats?

Yes, we are having issues with performance from a Database Server, taking a long time to run reports.

Server = HP Proliant DL380 with 2x Xeon X5675 Processors, 6 cores each @ 3 GHz and 98 GB Ram.  Storage is internal RAID 5 with 10k rpm drives.

There are 6 VMs on the Server.

[vMario156]

You need also to focus the VM itself and not only the esxi host.

How is the cpu and ram utilization of the VM durring you running your reports? if its fine check the same on the esxi host to see if there are other VMs may cause some ressource contention.

Next step would be the disk performance. Here you can get also some usefull informations via the graphs but also via the esxtop on the commandline.

Have you seen a better performance with the same configuration (number of VMs etc.) on this server?

Regards,

Mario

[CTSTech]

Here is the Setup:

Have 6 VMs:

1.  Database Server - new VM, Server 2008 R2, currently has 8 vCPU (2 sockets, 4 cores) and 32 GB Ram

2.  Terminal Server1 - was V2V'd from Hyper-V, Server 2003, currently has 2 vCPU and 4 GB Ram

3.  Terminal Server2 - was V2V'd from Hyper-V, Server 2003, currently has 4 vCPU and 4 GB Ram

4.  App Server - new VM, Server 2008 R2, currently has 1 vCPU and 8 GB Ram

5.  vCenter Mobile Access Server - currently has 1 vCPU and 512 MB Ram

6.  Managment Server - new VM, Handles Backups etc, currently has 3 vCPU and 8 GB Ram

The Database Server has plenty of resources and should be running the reports fast.

Disk in the graph shows:

1018 KBps average Usage

748 KBps average Write rate

270 KBps average Read rate

[vMario156]

Only the setup brings not much informations, because it really depends what these VMs are doing (load / IOs).

One tip in generall: If you are using a six core physical server you can may see a better performance if using 6 vCPUs instead of 8 (esxi will try to keep the load on one physical cpu, if using 8 it needs to be splitted up on both). This is of course not valid if your VM really utilizes more as 6 vCPUs at maximum level.

Regards,

Mario

[CTSTech]

vMario156 wrote:

Only the setup brings not much informations, because it really depends what these VMs are doing (load / IOs).

Which is why I want to know what Graph Counters are good to check and what numbers indicate good/bad performance

[harrygunter]

I would look at guest cpu ready time, you have guests with multiple vcpus on this host.

You may find dropping down the vcpu count actually increases performance.

Also disk latency is another one to look at, looks like you have plenty of ram on the host so shouldnt be a problem with ballooning.

[CTSTech]

harrygunter wrote:

I would look at guest cpu ready time, you have guests with multiple vcpus on this host.

You may find dropping down the vcpu count actually increases performance.

Also disk latency is another one to look at, looks like you have plenty of ram on the host so shouldnt be a problem with ballooning.

Thank you - I am looking for responses like this.

VM Ready Times:

1.  Overall - 134ms Average

2.  Database Server - 33ms Average

3.  Terminal Server1 - 32ms Average

4.  Management Server - 28ms Average

5.  App Server - 22ms Average

6. VMAS Server - 11ms Average

7.  Terminal Server2 - 5ms Average

Disk Latency - Highest Latency shows Maximum 1ms, Average 0.017ms

[harrygunter]

Are these averages taken when you are running reports or guest are being heavily loaded?

If not run the reports and look at the metrics then, that will giove you a better idea as to if there is a ready time problem.

Remember that when you look at the charts in the viclient they are in 20sec intervals, not realtime.

[CTSTech]

harrygunter wrote:

Are these averages taken when you are running reports or guest are being heavily loaded?

If not run the reports and look at the metrics then, that will giove you a better idea as to if there is a ready time problem.

Remember that when you look at the charts in the viclient they are in 20sec intervals, not realtime.

Those stats are just under normal use.  Do they look normal?

I will look at the stats when the report is being ran.

[MauroBonder]

Hi,

Read this DOC, its very good for this scenario - http://communities.vmware.com/docs/DOC-9279

also read http://www.yellow-bricks.com/esxtop/