analysis memory workload

[sorina]

Hi

I have some question? if any of this question is answred I will be happy.

1- why in memory metric is seperated to part: virtual and physical?

2- what is mapping betwen host memory-guest memory in vcenter and virtual memory-physical memory in vc Ops?

3-is  a document is pointed entitlement<=limit<=provisioned , so why my  "Anaytic VM " is not  follow this?

4- I did not do any reservation for my VM, but this view show me 206 MB reservation!!!!!!!!

5- why  amount of ram  is assigned to hypervisor?

Is there any technical paper for anaysis workload??? i have problem with concepts!!

thanks alot

[sorina]

Huuuraaa

question 1 is answred by kitcolbert . Thanks alot.

Other questions!!!!???

[admin · Accepted solution]

Hi Sorina

Some answers:

1. There is a diffirance in the way Memory is view from the Host and VM. Physical is what the VM is actually using (Memory metric : Active Memory).

2. vCenter is Consumed : vCops is Active memory

3.

4. The VM's memory overhead is always reserved and this is what you are seeing.

5. Are you looking for the Metric that is used to see how much memory is assigned to ESXi ?

[sorina]

thank a lot

in question 1 : memory in physical and virtual levels has demand , usage , entitlement  , what are diffrences three items in  both of them?

[kitcolbert]

sorina,

The differences in those metrics at the virtual level vs the physical level is due to ESX's memory reclamation techniques: transparent page sharing, ballooning, compression, and swapping.  The biggest differentiator is likely due to page sharing.  Where 2 or more virtual pages may be mapped to a single physical page.  In this way, the utilization at the physical level will be "less" than at the virtual level.  Hope that helps.

[sorina]

kitcolbert,

what about Demand? and entitlment ?  your answer was about usage that ,sometimes, one technique such as TSP is caused   usage of physyical level less than  usage of physyical level.

[sorina]

How can  you analyse this environment?

what statements are  derived from these view? one picture is virw of a VM and the othe is a  host.

do I need any other parameter (i mean parameters that are located in AllMetrics such as "ready time"  , ...)for checking and adjusting about enviroment?

[admin]

Hi Sorina,

From what I can see the VM does not have enought vCPU resourses. Looking at the config I assume you have 3ghz pCPU speed and that your VM only have 1vCPU. I would have added another vCPU to the VM....but only if you are having Issues with the VM Performance. (95% can be ok if nothing more is needed. )

The VM also have low contention.

Does that make sence ?

[sorina]

thanks , it 's ok . what about diffrence of demand memory in virt level and physic level ?

[kitcolbert]

Everything is to due to page sharing and the other memory reclamation techniques I discussed.  It doesn't matter which metric you use.

I've done many VMworld talks on this subject, I recommend you watch one or both:

- (from 2007): http://www.vmworld.com/docs/DOC-2116

- (from 2010): http://www.vmworld.com/docs/DOC-4643

[sorina]

thanks very much , But  before , I  had studied doc and conference of VMWorld . I need some doc or video about analysis health badge in operation manager  specially workload.

for example , in workload of memory in operation manager , there is  2 level "virtual" and "physical" and metrics such as entitlment,usage ,demand,... for both of them :

first ,Are  virt and phys levels from point  of OPs 's view  similar to levels that is pointed into VMWorld (Understanding Virtualization Memory Management Concepts-TA7750) (below picture)

If  NO , what is relationship  of  demand memory in physical and virtual level? a exmaple from usage parameter : usage in host  <=  usage in virtual leve (due to TSP)

******************************************************************************

second ,in  the other place of  this doc (TA7750) is pointed  "mem.consumed=  How much machine memory is allocated to running VMs and system services"

so  Should not cpu usage in operation manager is equal  with cpu consummed in vsphere ???? this definition says should be equal

I want you to help me.Is it possible? thanks

[sorina]

Why in this enviroment  usage memory in virtual level is hight while demand is very less. i mean that  why Hypervisor assign or allocate 1024 (100% confige) memory to vm ? is not it poor peformance . in other word why disallocate this memory from vm?

why vmware dose not have document for description of analysis workload of OPs???

thanks

[admin]

Regarding your questions around documentation on decription of analysis on vC Ops :

https://www.vmware.com/pdf/vcops-5-getting-started-guide.pdf P9

The Metric Concepts is decribed on P9.

Then you other question...why has the Hypervisor assign 1024Mb memory? You have configured the VM with 1024Mb memory...so the configured Capacity is 1024Mb.

[sorina]

thanks , first ,

HStrydom wrote:

Regarding your questions around documentation on decription of analysis on vC Ops :

https://www.vmware.com/pdf/vcops-5-getting-started-guide.pdf P9

  I have studied this doc , it shows some rules  such as : 1- Entitlement <= Limit <= Provisioned

2- Entitlement >= Reservations  BUT  this doc does not describe about relationship demand in virtual and Physical level , why sometime demand in physical (host)  is greater than virtual level ?

**************************************************************

seconde:  I mean that why usage=confige  while demand was low.

Then you other question...why has the Hypervisor assign 1024Mb memory? You have configured the VM with 1024Mb memory...so the configured Capacity is 1024Mb.

third , Is there any scenario or example of analysis in workload of vcenter operation manager?

[kitcolbert]

Ok, lots of posts to catch up.  First, the questions about my talk (TA7750).

1. Yes, the levels in my talk are the same as those in vC Ops.  I understand we use slightly different names, but it's the same thing.

2. You give the definition of memory consumed but then ask about cpu usage.  Did you mean to do that?  Assuming you did, yes, CPU usage and CPU consumed are the same.  Similarly, in the vSphere Client, memory usage is just memory consumed.  It's the same in vC Ops.  (vC Ops just also shows demand alongside usage.)

[kitcolbert]

Sorina,

To your next question, the reason that the hypervisor has allocated 1024 of virtual memory to the VM is because the VM has allocated that much memory and because the host is not memory overcommitted.  As I mentioned in my talk TA7750, ESX will only reclaim memory from a VM if another VM needs it.  In your case, there is enough host memory to satisfy all VMs' memory allocations, meaning that the host will never have to reclaim memory from a VM.  I understand it may be confusing as to why the host doesn't take memory back if a VM isn't "using" it, but the reality is that memory reclamation has a performance cost (generally minimal) that we like to avoid if possible.

[kitcolbert]

To your final questions:

1. Physical demand can be both lower and higher than virtual demand:

     - It can be lower because of transparent page sharing and other reclamation techniques.  With TPS, you're essentially deduplicating memory, leading to less on the physical level than on the virtual level.

     - It can be greater due to the fact that ESX requires additional "overhead" memory in order to run the VM.  This memory only exists on the physical level and not on the virtual.  So if the VM has no page sharing and no other reclamation, then once you add in overhead, the physical level can be greater than the virtual level.

2. I answered this question in my last post.

3. It's a good suggestion.  We don't have one now.  I can look into whether we can get some examples added.

[sorina]

kitcolbert,

Thank you very much, I have learn many things from these post.and your docs

may I ask another questions??

  Our firm is using trial vcenter operation manager ,SO in first phase of project we have worked health badge and workload (minor badge).we want to know how a non-expert man can analyze from monitoring vcenter operation manager. (in vcenter it was hard)

1) is it enough  to check  cpu , memory,storage and network workload  ?  why in some posts  is pointed to check  Ready and Contention Graphs (I know what these are) , I mean that when we  should check these graphs (just in status that demand>usage) . please explain  just a example

  1-1) what  other metrics  should be checked?( in Allmetric is many metrics)

  1-2) Is it possible  some of  important metrics such as balloon memory  is interpreted in  view of workload in vcenter operation manager ? (i know no direct mapping betwen vcenter metric and vps )

  1-3)  Is it necessary both of ready time and contention is checked  in status that demand>usage or one of  them is enough , I think that when we have cpu ready time (above 5% , i  see in some weblogs) we should understand we have cpu contention .

2) how much contention or ready time is normal?