VMware Cloud Community
zenomorph
Contributor
Contributor
Jump to solution

VMware ESXi scale up vs. scale out

Hi there........

Were reviewing our VMware practice and would like to know the general practice out there in VMware Cluster implementation - is it more common to Scale-Out or Scale-Up the servers eg. use 4-Way Servers vs. 2-Way Servers?

Obviously 4-Way will provide greater level of consolidation and VM density due to vCPU count while 2-Way will be lower but are there any other arguments eg. placing less of your eggs in one basket. From a cost perspective it would seem 4-Way will be more cost effective.

Any comments..

Cheers........

0 Kudos
1 Solution

Accepted Solutions
markdjones82
Expert
Expert
Jump to solution

I agree with Philly.  We prefer the 2 socket setup and scaling out, preferably blades as well.  If you are still doing rack mount I would highly suggest looking into blades.

Some people do swear by the 4 socket, but you usually hit memory limits way before you will hit CPU limits.

http://www.twitter.com/markdjones82 | http://nutzandbolts.wordpress.com

View solution in original post

0 Kudos
5 Replies
JCMorrissey
Expert
Expert
Jump to solution

Hi,

Just checking, might worth looking at the attached thread about the very same discussion. Fairly detailed and covers a multitude of angles

http://communities.vmware.com/message/487409#487409

Hope this helps!

Please consider marking as "helpful", if you find this post useful. Thanks!... http://johncmorrissey.wordpress.com/
PhillyDubs
Enthusiast
Enthusiast
Jump to solution

You've pretty much figured it out. With the way licensing is going on with VMware(specifically the vCloud Suite), the strategy I find worthwhile is utilizing two socket systems with 256 or 384GB of RAM if CPU is not your limiting resource. This still allows high density, but not so much as throwing twice as many virtual machines on a four socket 512GB of RAM system. I've been down the road of both setups and the four socket systems were pure hell when the host hardware had issue. Without 10GbE at the time, maintenance was also a pain.

VCP5
markdjones82
Expert
Expert
Jump to solution

I agree with Philly.  We prefer the 2 socket setup and scaling out, preferably blades as well.  If you are still doing rack mount I would highly suggest looking into blades.

Some people do swear by the 4 socket, but you usually hit memory limits way before you will hit CPU limits.

http://www.twitter.com/markdjones82 | http://nutzandbolts.wordpress.com
0 Kudos
WessexFan
Hot Shot
Hot Shot
Jump to solution

"Obviously 4-Way will provide greater level of consolidation and VM density due to vCPU count while 2-Way will be lower but are there any other arguments eg. placing less of your eggs in one basket. From a cost perspective it would seem 4-Way will be more cost effective."


This isn't altogether true. As Mark states, today's processors are so powerful that you can get great performance with 2 sockets and 8-16 cores and you will hit vRAM contention way before CPU resources are touched.

VCP5-DCV, CCNA Data Center
0 Kudos
MKguy
Virtuoso
Virtuoso
Jump to solution

It should also be noted that in terms of raw CPU performance, due to architectural reasons a single core of a 4-socket system is usually more inefficient compared to a core of a 2-way CPU family based processor, even at identical clock speeds (at least in the current Intel-based world):

http://www.spec.org/cpu2006/results/res2013q2/ (I'm aware this may not provide perfect apples-to-apples comparisons, but you should get the idea when comparing similar setups)

Take that into account and consider that a 4-socket based (Intel) CPU comes at almost twice the cost (apart from the higher cost of the base-system).

So in terms of raw CPU performance and capex, two 2-socket (Intel) servers should usually beat a single (Intel) 4-socket clearly.

-- http://alpacapowered.wordpress.com
0 Kudos