Last month at VMworld, VMware announced the upcoming release of their virtualization suite, vSphere 5.5. As you all know, I am constantly watching the hypervisor space and work to determine how each new feature can benefit database professionals. With every release of the major hypervisors lately, the limits of the technologies continue to be stretched, and new features pop up and surprise us all. This release is no different, and I am very excited about some new technologies that benefit us all. Let’s dive in!
First, for the formal announcement, VMware has a “What’s New†PDF available for you to review here.
Maximums
Each release of the hypervisors always seem to push the maximums list bigger and bigger. First and foremost, each individual virtual disk, or VMDK, now has a maximum size limit of 62TB, up from 2TB. That’s fantastic! It also upped the RDM limit to this maximum as well. Up to 120 virtual disk and CD-ROM devices can now be connected to each individual VM. Do the math and you find that you can now present 7.44PB of storage to an individual VM. Nice!
Each host now supports 320 physical CPU cores per host, with 4TB of memory and 16 NUMA nodes. 16Gb fibrechannel adapters are fully supported, and 40Gb network adapters are now supported.
As a result of these improvements, I would really have to struggle to find a physical machine that I would not recommend virtualizing at this point. In the past, it was a very rare occasion to find something that was not suitable for virtualization, and it was always the hardware limitations. At this point, those limits are flat-out removed.
Clustering
Microsoft Clustering continues to be supported with vSphere 5.5. Windows Server 2012 is now explicitly supported, as well as using iSCSI and FCoE for shared storage, as well as round-robin storage multi-pathing. Up to a five-node cluster is supported.
VM Latency Sensitivity
SQL Server, and all other DBMS applications, are very latency sensitive when handling resource waits. Rather than go through a lot of individual settings that VMware published for low-latency requirements for critical systems, vSphere 5.5 now has a setting that does all of this for you. It helps to prioritize resource contention, reserving memory, monitoring CPU dedication, and eliminates some network features that can induce latency delays in the stack. This feature greatly simplifies my SQL Server VM construction process. The VMware Vroom! blog has more details on this feature here.
vSphere Flash Read Cache
This feature is incredible. Most SQL Server workloads are very read intensive, and caching these requests can dramatically reduce the I/O requirements of an individual server, while improving the overall performance. You can also reduce the amount of memory allocated to the SQL Server if the workload allows for it and the caching is configured properly. As a result, you can increase your SQL Server virtualization consolidation ratio without sacrificing performance, which should save you in SQL Server licensing costs.
Previously, I would recommend host-based SSD caching for major SQL Server workloads that were heavy in their storage read requirements. Third-party vendors have stepped up and offer some great solutions for this need.
Now, VMware has added this functionality into the core stack. If you have SSD devices in the physical host servers, you can enable this feature quickly and easily. Once the cache underneath the SQL Server VM warms up, the performance improvements to that VM can be huge.
Application High Availability
vSphere 5.5 now comes with a service-level awareness feature via vFabric Hyperic that can monitor certain services for their runtime state. SQL Server is line-itemed as one of the first supported application. If a failure of the service is detected, this feature can be configured to restart the VM after a specified period of time. The goal is to improve the high availability of the service without the need for additional hardware or licenses.
Big Data
I know, I hate the ‘big data’ buzzwords floating around right now, because everyone’s definition of ‘big data’ is different… It’s just like everyone’s definition of ‘cloud’, right? But, if you are running Hadoop, vSphere has added some “Big Data Extensionsâ€, or BDE, that can help you manage Hadoop clusters. It supports all of the major Hadoop distros.
Conclusion
I am personally very excited for this upcoming release, and as soon as the released date is announced, I’ll post it here! These new enhancements and features can be utilized to improve our system performance and high availability levels.
Check back soon for a similar post describing all of the exciting updates that come with the recent announcement of Microsoft Hyper-V 2012 R2.
Hello,
I’m implementing vFRC on our new 5.5 hosts. I unfortunately manage our SQL servers as well. Have you experimented with local read caching on SQL server yet? I’m assuming data partitions can best make use of this technology. I’m unsure how much cache I should allocate for a given partition. My hosts have 100G SAS connected SSD’s. VMDK’s sit on FC connected VNX LUN’s with fast cache.
If I had to take a stab is vFRC config on a 300G data VMDK I might go for a cache of 10% at 30G using 64K block size… but that is pulling it of my hind side.
Thoughts?
Again, this is another great question! I’ve done a little bit with local SSD read caching, and am a huge fan. If you have 300GB of data on a VMDK, what percentage of that VMDK is changing over the course of a day or week? How much of the I/O is read and how much is write? Glenn Berry’s diagnostic queries, located at sqlserverperformance.wordpress.com should be able to help answer these questions.
Off the top of my head, a 10% buffer at 30GB sounds good, but I’d use 8KB blocks as opposed to 64KB blocks, since SQL Server reads data files in 8KB chunks. Let me know and we’ll go from there!
It’s been a while since I left that comment regarding vFRC. As it turns out vflash in general is broken and may not ever be fixed in 5.5.x. I’ve had a couple cases with vmware on vcenter issues and every time they look to see if vflash is enabled on any hosts and if so disable it. My SSD’s are so sad.
Eeeek! This is the first I’ve heard of any negatives about VFRC. Can you post any links that support your comments?