VMware is a company known for revolutionary change. It was not many years ago that virtualization was nothing more than a toy used by developers and IT enthusiast on their desktops. Today it is the heart of datacenter 3.0 and the only viable platform for cloud computing.
With the release of vSphere 4.1 VMware is maturing into a company that not only is able to stir up a revolution but is then able to establish a nation. This release is not as much of a revolution beyond the previous 4.0 product but more of an evolution. As VMware consistently delivers minor releases that not only offer fixes but a solid plethora of new features we can be assured that we are on a steady upward climb of quality and quantity.
Some of the major differences found in VMware vSphere 4.1 are specific aimed toward technologies that are transforming IT operations. One of these is the Virtual Desktop revolution that VMware is spearheading. Memory Compression is a new hierarchy for VMware’s memory overcommit technology. It is a new level of the memory hierarchy, between RAM and disk. Slower than memory, but much faster than disk, this feature improves the performance of virtual machines when memory is under contention, because less virtual memory is swapped to disk as a result. This has the net effect of greatly improving Virtual Desktop workloads because RAM memory is the key resources needed to drive this new workload.
Other changes in datacenter design such as 10GbE are also changing the design critera for vSphere deployments. With traditional 1GbE design many NICs were used and traffic was easily able to be separated at a physical layer. With the introduction and mainstreaming of 10GbE it is now impartitave that we have granular control over how the much larger pipes are used. In most 10GbE VMware ESX server designs only 2 10GbE NICs are used. This gives the advantage of a much larger pipe but lacks the physical separation layer needed to properly tune network traffic. So long as the load offered to the 10GbE interfaces is less than 10GbE, everything is ok—the NIC can service the offered load. But what happens when the offered load from the various traffic types exceeds the capacity of the interface? What happens when you offer 11Gbps to a 10GigE interface? Something has to suffer. This is where Network IO Control release as part of vSphere 4.1 steps in. It addresses the issue of oversubscription by allowing you to set the relative importance of predetermined traffic types.
Load Based Teaming is yet another new technology introduced in vSphere 4.1. Consider the situation where we have 10 VMs allocated to a NIC team with two NICs using Originating Virtual Port ID. Five VMs will use one vmnic, and the other five will use the other. The allocation of the VMs is independent of their network I/O load. If one or two VMs allocated to one vmnic are quite traffic intensive, then we might get an imbalance and congestion on one vmnic. Load Based Teaming seeks to solve this by moving reallocating virtual ports to vmnics when congestion is detected on a vmnic. Congestion is signaled by transmit or receive traffic exceeding a 75% mean over a 30 second period.
Much attention has been paid to the networking portion of the vSphere 4.1 release. Among the primary features just mentioned we also have vmkernel improvments for vMotion, NFS, and FT logging. UDP and intra-host VM to VM performance is also improved with vDS scaling to ~350 hosts (64 is the current limit).
It does seem as if storage and VMware have a love hate relationship. VMware relies of advanced storage arrays from manufactures such as Netapp and EMC yet when there are VM performance problems many times it is the interaction with how this storage is utilized by VMware that is causing it. The problem Storage I/O control is addressing is the situation where some less important workloads are taking the majority of I/O bandwidth from more important applications. If we are given a scenario of 3 VMs on one ESX server a data mining server, an Exchange server and an online store application. It is very possible that the data mining, the least business critical application, is consuming most of the storage I/O resources. What we want to see is a distribution of I/O that is aligned with the importance of each virtual machine. Where the most important business critical applications are getting the I/O bandwidth needed for them to be responsive and the less critical data mining application is taking less I/O bandwidth. Storage I/O control is able to deliver this.
vSphere 4.1 is also a continuation of the push toward ESXi hypervisor layer without the ESX service console. ESXi is continuing to evolve into the primary role vs. its current secondary role as the VMware hypervisor. Many new features have been included such as:
- New Deployment Options
- Boot from SAN
- Scripted Installation (a la “Kickstart”)
- Centralized updating of 3rd party code with Update Manager VMware Update Manager can deploy drivers, CIM providers, other modules
- Improved Local Authentication
- Built-in Active Directory Service
- DCUI and Tech Support Mode access by any authorized user (not just root)
- Easier CLI options for troubleshooting
- Full support of Tech Support Mode – both local and remote (via SSH)
- Additional commands in Tech Support Mode: vscsiStats, nc, tcpdump-uw, etc.
- Additional management options in vCLI: SCSI, VAAI, Network, VM
- Better control over local activity
- DCUI and Tech Support Mode is configurable in vCenter Server
- Total host lockdown possible
- Activity in Tech Support Mode is sent to syslog
In the future major releases of VMware vSphere will include only the VMware ESXi architecture.
Other improvements to vSphere 4.1 include:
DRS Host Affinity which is the ability to restrict placement of a virtual machine to a
subset of host in a cluster.
DPM Scheduling is a needed feature of Dynamic Power Management. It gives the ability to schedule when DPM is engaged and when it is not. DPM has had limited adoption because of some of the side effects of it being on at all times.
vSphere 4.1 is the continual improvement of an already great product. The improvements in established features and the addition of new ones combine to make a platform that is the choice for building a tangible Cloud architecture.
