It is really hard to write about vSphere products more interesting and intelligent that Duncan Epping does, however, I am so excited about all I have heard and read today (and I am sure I will read more information for at least next 2 hours), so I decided to publish some short notes I could catch in a waterfall of vSphere 5 new features.
- Currently about 40% of server workload now is running in virtual environment, this number will reach 50% by the end of 2011
- The main point of first part of the presentation was mostly the global move to the cloud computing model - either private or public. Now it is not only virtualizing your physical servers, but providing Virtual Machines AppStore to the clients. Basicallly, VMware is pushing all vSphere’s implementations to IT As A Service model.
- New cloud infrastructure suite now consists of the following products:
- vSphere 5
- vCenter SRM 5
- vCenter operations 1.0
- vShield Security 5
- vCloud Director 1.5
My short experience covers mostly vSphere, so I tried to write down everything about it, and therefore I missed other products’ new features.
Let’s concentrate on the most important part of
vSphere 5 improvements:
- Profile Driven Storage – allows you to create storage tiers or they call it also Datastore cluster. You create for instance 3 different performance specs datastore clusters using flash disks for Tier 1, SAS disks for Tier 2 and FATA disks for Tier 3. When you create VM you just assign it to the proper Datastore Cluster according to the service level requirements of VM. If performance of current VMFS doesn’t meet VM’s requirements the VM will be migrated to better VMFS datastore. To me it seems like a software replacement of the multi tier hardware storage solutions like HP 3Par storages.
- Storage DRS – it is very close to regular DRS. It will also take care of initial placement of VM with regard to available space, you can create your affinity rules for keeping some of the VMs separate across different datastores, it will migrate your VMs according to IO balancing level you set and space allocation rules you create. Finally, you can move your datastore to Maintenance mode so all VMs will be moved to other datastores for the period of maintenance.
- VMFS 5:
- All VMFS datastores are formatted with standard 1MB blocksize
- The new VMFS can grow up to 64TB
- VMDK file is limited by 2 TB
- vSphere 5 support new and old versions of VMFS
- If you want to upgrade to VMFS 5 it is easy and fast to do
- vSphere Storage Appliance – not sure if I got all details correctly. Briefly, it will let you use all local ESXi disks to create kind of virtual shared storage. You don’t need shared storage, but you still can use some benefits of it. But there was nothing told about its limitations and restrictions – something to investigate for me. I guess it is mostly aimed for small vSphere implementations.
- New Hardware Version 8, including 3D graphics and MacOS X server support
- New VM’s maximums - now you can create Monster VM:
- 32 vCPUs
- 1TB of RAM
- 36 Gbps
- 1,000,000 IOPS
- High Availability
- The concept of HA has been changed completely to Master/Slave model with an automated election process. There is now only one Master and the rest of nodes are slaved. Master coordinates all HA actions. With this new model vSphere admins don’t need to worry about HA Active hosts placement and distribution accross blade enclosures
- HA doesn’t rely on DNS anymore
- The big change for HA is that now all nodes will use storage paths and subsystems for communication in addition to network. This will help nodes better understand the health of their neighbor nodes.
- FT VMs still have 1 CPU limit, but probably we can use more than 1 vCore per vCPU with FT VM - have to check. VMware just increased range of FT supported CPUs and Operating Systems.
- Auto Deploy – it is a tool for easy bulk deployment of ESXi hosts. vCenter can keep Image and Host profiles. You can create some rules that will tell vCenter what Image, Host profiles and Cluster to use while installing new host. The bad thing for me about it is that I need to start learning PowerCLI – you need these skills to create auto deployment rules.
- Enhanced Network IO Control – Now it is per Vritual Machine control. In vSphere 4.1 it was per port with vDS.
- Storage IO Control – Added control for NFS storages
- vMotion – finally VMware implemented load balancing of vMotion over several vmotion enabled VMKnics. vMotion is now supported with latency up to 10ms.
- vCenter linux based Virtual Appliance – you can still use vCenter on Windows. Didn’t get what the difference is between those two
- Vcenter SRM 5
- No more requirement for hardware array based replication. It is all done now in software. You can have different storage in your Disaster Recovery site
- Failed VMs can fallback to the main vSphere once it is fully restored
- Proactive migration using SRM – for instance if you expect power outage in your main datastore you can manually initiate failover to DR site
- vCenter operations 1.0 - SLA Monitoring
Here comes the biggest concern of all admins that already built some plans for upgrade to vSphere 5 – licensing!
The licensing is still counted in CPU units. If you have 4 CPUs in 2 ESXi hostss you will ned 4 licenses. The good news is that your CPUs are not limited by number of cores anymore. Another good news is elimination of RAM limit per host.
Now we can proceed with bad news. VMware creates new definition – vRam Entitlement, that is, how much RAM you can assign to your VMs per license. For instance, with 1 license for vSphere 5 Enterprise edition you are entitled to use 32GB of RAM. Even if your host has 48GB of RAM you won’t be able to assign to your virtual machines more than 32GB. You can merge vRAM entitlements into Pooled vRAM, that is, the sum of your vRAM entitlements across all hosts connected to your vCenter or across all linked vCenter instances. Consumed vRAM has to be less than pooled vRAM. Amount of entitled vRAMs differs across different vSphere editions.
Here is another example of pooling vRAM. You have 2 hosts with 2 CPUs and 96 GB of RAM each. With 4 Enterprise Edition licenses your pooled vRAM equals to 128 GB, but you want to use all 192 GB your hosts have. Then you will need to buy 2 more Enterprise edition licenses. Even though you don’t have two more CPU you can use vRAM entitlement that comes together with license per CPU.
As I understood switched off and suspended VMs are not counted in pooled vRAM. I think it will significantly change approach to VM’s memory provisioning and management. I can imagine that some admins will start providing VMs with less RAM than they need trying to cut licensing expenses.
Those companies that have high overcommitment memory will need to pay extra for upgrade to vSphere 5, which makes all memory overcommitment technologies less valuable now. No matter how many GBs you can save using Transparent Page Sharing – you still have to pay for all allocated vRAM.
There is also tool that can help you to calculate your consumed amount of vRAM and give you an idea on how many licenses to buy and what edition to choose.
For owners of Enterprise License Agreement the upgrade process is quite straight forward – just contact VMware, they will get provide licenses to you.
Seems like with new licensing model VMware is more aimed at the market of big companies, where flexibility of pooled vRAM across multiple vCenter instances can save some money. SMB companies will definitely need to pay more for vSphere 5.
Update 1: I should never use MS Word again for blogging. Screwed all my text formatiing.
Update 2: Guys, I will really appreaciate your feedback and comments about the blog content's quality.
Update 3: There is no firm restriction on vRAM allocation for Enterprise and Enterprise Plus editions. If you use more vRAM than you are enttitled for you will be given a warning.
If you find this post useful please share it with any of the buttons below.
