How much ram do you allocate for a vm?

tdean

I dont think our terminal server vm's are performing as well as they should. I was on the vmware boards and they claim you should almost never allocate more than 512mb ram for a windows server vm. i looked at ours and they have over 2.4Gb allocated. at some point, i will be upgrading this from ESX3.x to ESXi 4.1. when i do that, should i change the ram settings?

certhelp

tdean wrote: »

I dont think our terminal server vm's are performing as well as they should. I was on the vmware boards and they claim you should almost never allocate more than 512mb ram for a windows server vm. i looked at ours and they have over 2.4Gb allocated. at some point, i will be upgrading this from ESX3.x to ESXi 4.1. when i do that, should i change the ram settings?

512mb RAM? Even with 1024MB/1GB RAM, Server 2008 can crawl. Does Your host on which the VM is hosted have enough RAM? It should have enough RAM for all the guests + the host. I don't think 512MB is max RAM is a valid advice.

gateway

I am far from a VMWare expert, but that advice doesn't sound correct to me. We have 2gb for a lot of our virtual servers. Have you done any performance monitoring of the servers? What do the ESXI logs say? I'm sure someone here will be along very shortly to clear this one up

tdean

certhelp wrote: »

512mb RAM? Even with 1024MB/1GB RAM, Server 2008 crawls. Does Your host on which the VM is hosted have enough RAM? It should have enough RAM for all the guests + the host. I don't think 512MB is max RAM is a valid advice.

well, im new to this so i'll do my best. the physical machines have 24Gb ram. it looks like vmotion, ha and drs are set up on the cluster (3 physical servers) and 10 vms are on 1 server, then our 2 Term servers are each on their own box (one has the file server) and the host cpu/mem is pinned all the way across.

heres the link to the vmware thread:
http://communities.vmware.com/thread/74184

SteveO86

Yea.. 512 MB for a server.. That's..not....a solution... but hey one of those guys are a vExpert... So maybe it's me.

Are you sure the host only 2.5 GB of RAM, I would definitely get that upgraded.

tdean

SteveO86 wrote: »

Yea.. 512 MB for a server.. That's..not....a solution... but hey one of those guys are a vExpert... So maybe it's me.

Are you sure the host only 2.5 GB of RAM, I would definitely get that upgraded.

each physical machine has 24Gb.

Essendon

I aint a VMware expert either, but I can say that 512MB RAM for a terminal server is not enough in many if not most cases. How many users jump on them at one time? What software do they use on the terminal servers? If people were using say for example AutoCAD, you require much more RAM than 512MB. Otherwise, 512MB RAM is enough for Windows Server 2003 DC/DNS/DHCP server in a small(ish) environment.

Essendon

Devilsbane wrote: »

I like to do 1GB if I can, but 512MB is normal and I have even ran on 256 before.

At least with my machines, I'm really not doing anything on them. They usually just sit there idle until I go test something, so there really isn't a large demand for performance. These are also running xp/2003 and not a memory hog like Vista.

I think he's referring to his prod environment.

tdean

Essendon wrote: »

I think he's referring to his prod environment.

yeah.... and the TS users are running an app suite called NextGen. I think its odd that 1 server has 1 vm, 1 server has 2 vm's and the 3rd has 12. this is with hs/drs enabled in 3.5.

im just starting in on this stuff now, and my class isnt for a few weeks.

MentholMoose

The thread you linked is ancient so don't pay attention to it. The advice was OK at that time since RAM was more expensive and people were more conservative with allocating it to VMs. Also at least one reply was clearly recommending 512MB as a starting point... generally you give VMs fewer resources and only allocate more if you have a reason to. If you notice in that thread, the OP allocated 2GB RAM to a VM when the host only had 2.5GB of physical RAM, which may not leave enough for the host (let alone other VMs).

If you are having performance problems in a VM, check the resource utilization for that VM. You can do that by just treating it as a physical server and using tools like Task Manager or Process Explorer inside the VM. In your screenshot, it looks like one of the VMs is alerting for high CPU usage and/or high RAM utilization, so look into what is causing that. Also, what type of datastore are these VMs stored on (local, iSCSI, FC, NFS)?

tdean

MentholMoose wrote: »

The thread you linked is ancient so don't pay attention to it. The advice was OK at that time since RAM was more expensive and people were more conservative with allocating it to VMs. Also at least one reply was clearly recommending 512MB as a starting point... generally you give VMs fewer resources and only allocate more if you have a reason to. If you notice in that thread, the OP allocated 2GB RAM to a VM when the host only had 2.5GB of physical RAM, which may not leave enough for the host (let alone other VMs).

If you are having performance problems in a VM, check the resource utilization for that VM. You can do that by just treating it as a physical server and using tools like Task Manager or Process Explorer inside the VM. In your screenshot, it looks like one of the VMs is alerting for high CPU usage and/or high RAM utilization, so look into what is causing that. Also, what type of datastore are these VMs stored on (local, iSCSI, FC, NFS)?

Hi MM,
i didnt concider ram prices and the age of that thread. Well, that takes ram out of the equation. I cant seem to figure out why i have that alert. i went in like a "regular" server and saw the performance monitor says its using ~5Gb page file and 10-20% cpu. we have 3 vm boxes connected to an AX4 with fiber.

MentholMoose

You can check the Alarms tab in vSphere/VI Client to get more info on the alarm. In your screenshot it shows 4260MHz CPU utilization for one of the VMs. What is it now, and how many vCPUs are allocated to the VM? This will make a difference... if for example there are two vCPUs and the CPU speed is 2.2GHz, then when you took the screenshot the VM was at about 100% CPU (2130MHz * 2) which could generate an alarm.

Also check the Performance tab for the VM in vSphere Client. You can check the current state and history for various resources (CPU, RAM, disk, network, etc.). It may not be at 100% CPU now, but maybe it was earlier, and you can check when. Or maybe another resource was over-utilized. Check the graphs and compare to when you had slowness complaints.

You can also check the Performance tab for the hosts. A particular host could be overloaded at certain times. You can try vMotioning some of the VMs to other hosts to redistribute the load. However it sounds like you are saying the two TS VMs are on their own hosts and they are maxing out the host resources. In this case migrating them to another host wouldn't help and you need to investigate why the VM utilization is so high. This is not a VMware thing, rather it is some app in the VM using a lot of resources, so troubleshoot it like you would a physical machine.

blargoe

From the screen shot it would seem that CPU is a possible culprit. You can go to the performance tab of the VM and look at the charts for CPU, Memory, Disk, etc and get a look at a day or two's worth of performance trends right from the VMware client. I'd be curious to see if you're getting spikes on the CPU% or CPU Ready time for this VM.

Do you have any more physical resources that can be added to the VM? Any resource pools/reservations/limits in play here?

tdean

thanks for the info guys... i will check this tomorrow and let you know.

EDIT: ok, i vpn' d in. the perf tab says the cpu usage is very high. i looked and 4 vCPU's are allocated to this vm. i logged in to the server itself and cpu usage was below 3%.
it has been like this since we had a power outage last week and everything crashed.

blargoe

This might sound crazy but maybe you have too MANY vCPU's assigned to the VM. When you have multiple vCPU's, the VM's OS will not process any instructions until all 4 cores are free at the same time (this is the "CPU Ready" metric that you will find in the performance tab for the VM).

tdean

i dont know... i think when i upgrade to esxi 4.1 im just going to have to completely reconfig all these. the 3 physical servers are identical, and the other term server is identical to the one giving alerts and it is not giving alerts, however seems to be high cpu/mem as well.

MentholMoose

How many cores do the hosts have? If you can schedule a short maintenance window, you can try reducing the vCPU count for one of the problem VMs to see if it resolves the issue. The procedure is simply 1) shutdown VM, 2) edit vCPU count in VM Properties, and 3) boot VM.

tdean

MentholMoose wrote: »

How many cores do the hosts have? If you can schedule a short maintenance window, you can try reducing the vCPU count for one of the problem VMs to see if it resolves the issue. The procedure is simply 1) shutdown VM, 2) edit vCPU count in VM Properties, and 3) boot VM.

i think they are 4 quad core cpus. they are proliant dl 350's??? i think..... i will check when i get in there tomorrow. i'll post more info and maybe we can come up with a plan to get these working properly if you guys dont mind.