3 ms·

Just a reminder, adding VMWare-like virtualization layer (i.e. full virtualization) to the mix will cost you more than 40% of your CPU: http://domino.research.i

Just a reminder, adding VMWare-like virtualization layer (i.e. full virtualization) to the mix will cost you more than 40% of your CPU: http://domino.research.ibm.com/library/cyberdig.nsf/papers/0929052195DD819C85257D2300681E7B/$File/rc25482.pdf http://domino.research.ibm.com/library/cyberdig.nsf/papers/0... In my view one of the key reasons to do it is to use AWS or the like. Until they have full Docker support on bare metal that is. P.S. There are other reasons, e.g. Docker makes it difficult to create a separate publicly visible network interface. But these I feel will go away soon as well.

skybrian 12y ago
Looks like VMWare is working on that too: http://googlecloudplatform.blogspot.com/2014/08/containers-vms-kubernetes-and-vmware.html http://googlecloudplatform.blogspot.com/2014/08/containers-v...
acdha 12y ago
> Just a reminder, adding VMWare-like virtualization layer (i.e. full virtualization) to the mix will cost you more than 40% of your CPU … on a couple of microbenchmarks when you define “CPU” to mean “I/O”. Even if you ignore the question of whether KVM performance is the same as VMware's (hint: no), most of the the charts in that paper contradict such a broad statement.
jasongill 12y ago
This article only mentions VMware twice in the body, both in passing. The majority of the article is a comparison of KVM and Docker. Many of the points in the article don't apply to VMware at all, or refer to things that are KVM specific limitations which VMware has already overcome. In fact, the article says "KVM has much higher overhead, higher than 40% in all measured cases." - saying nothing of "VMware-like virtualization" or VMware ESXi at all. While it's true that containerized "virtualization" traditionally has less overhead than a hypervisor like ESXi, the difference is increasingly small and in most environments negligible, especially considering the added features and flexibility of ESXi vs containers. From my experience and tests, the performance difference between an app in a container running on bare metal vs the same app running in VMware has been insignificant, so to say that there is a 40% performance penalty is probably disingenuous.

skybrian 12y ago

Looks like VMWare is working on that too: http://googlecloudplatform.blogspot.com/2014/08/containers-vms-kubernetes-and-vmware.html http://googlecloudplatform.blogspot.com/2014/08/containers-v...

acdha 12y ago

> Just a reminder, adding VMWare-like virtualization layer (i.e. full virtualization) to the mix will cost you more than 40% of your CPU … on a couple of microbenchmarks when you define “CPU” to mean “I/O”. Even if you ignore the question of whether KVM performance is the same as VMware's (hint: no), most of the the charts in that paper contradict such a broad statement.

jasongill 12y ago

This article only mentions VMware twice in the body, both in passing. The majority of the article is a comparison of KVM and Docker. Many of the points in the article don't apply to VMware at all, or refer to things that are KVM specific limitations which VMware has already overcome. In fact, the article says "KVM has much higher overhead, higher than 40% in all measured cases." - saying nothing of "VMware-like virtualization" or VMware ESXi at all. While it's true that containerized "virtualization" traditionally has less overhead than a hypervisor like ESXi, the difference is increasingly small and in most environments negligible, especially considering the added features and flexibility of ESXi vs containers. From my experience and tests, the performance difference between an app in a container running on bare metal vs the same app running in VMware has been insignificant, so to say that there is a 40% performance penalty is probably disingenuous.