Supercharging VMware Infrastructure
With #VMworld 2018 kicking off in Las Vegas this week, this is an opportune time to look at the core requirements that accelerated the growth of VMware virtualization in the last decade and what to expect in the months following VMworld. Working closely with customers that are driving new levels of application integration, the team at Vexata has seen first-hand, the benefits of using a NVMe-based storage platform to support large-scale virtualization initiatives. When executed correctly, these application platforms deliver measurable business benefits by delivering the performance to achieve higher levels of scale to enable massive consolidation, cost savings and operational efficiencies.
Today’s virtual server infrastructure has been battle tested and platforms such as containers and cloud that were initially viewed as a threat to VM architectures, actually represents an opportunity for continued innovation as represented by VMware’s investments in these areas. It’s critical for VMware to continue to expand into these markets, as they still account for the majority virtual server deployments.
Analyst studies show that on an average each x86 server runs 12 VMs – that translates to 100’s of VMs per datacenter rack. Given that a single multi-core server can drive more than 20GB/s of throughput, this type of traffic loading can easily saturate existing AFA architectures, forcing IT architects to buy more AFAs to distribute the load and scale performance. Modern application workloads are driving higher levels of bandwidth, IOPS and faster response times, further exposing the storage controller bottlenecks that exist in almost every AFA architecture.
How can architects overcome these storage architecture limitations?
While Moore’s law is definitely showing signs of a slowdown, Intel X86 platforms continue to push performance boundaries every year. The Xeon Scalable Platform offers up to 112 virtual CPUs in a typical dual socket server. A number of databases VMs can be easily consolidated on each server, but only if there is a low latency storage infrastructure driving the I/O. For high performance virtualized server use cases, the storage infrastructure must be able to deliver more IOPS, mixed read/write throughput and lower sustained latency under load.
Let’s take the example of Oracle transactional and analytic workloads being virtualized within a VMware ESX infrastructure. The ability to consolidate the workloads from multiple OLTP and OLAP hosts across a virtualized compute environment inevitably leads to the unfortunate discussion about performance tradeoffs, even with AFAs. NOT WITH VEXATA!
Why Vexata for virtualized workloads?
When faced with the objective of running production databases, analytics and reporting on virtual infrastructure, experienced architects know how to design systems that simultaneously run these extreme mixed I/O workloads without hitting performance rails. Within the domain of the VMware ESX infrastructure, scaling is as simple as spinning up more VMs on the fly as the workloads demands increase. This on-demand scaling is matched within the Vexata architecture, and by utilizing the in-box scaling and analytics, architects can non-disruptively and cost-effectively scale performance and capacity together.
Figure 1: VX-100F In-Chassis Scaling
To add VMs with existing AFA architectures, the admins are forced to find more rack space, floor tiles, power, cooling, with the network connections and cabling mess to add into the cluster. With Vexata it is very simple – the customer just adds more Enterprise Storage Modules (ESMs) to their existing array without needing to add additional IO controllers within the same 6U chassis.
Figure 2: Consolidation, scale and performance for VMware
Vexata has recently performed a set of benchmark tests with the VX-100F using multiple “heavy” transactional and analytical workloads virtualized with VMware ESX 6.5. This was demonstrated using 6 “heavy” VMs using 3 OLTP workloads, a Web application and 2 separate analytic workloads. These benchmarks demonstrated that with Vexata, these VMs delivered bare metal performance under ESX management.
This clearly illustrates that you can consolidate intensive workloads without performance degradation, delivering real-time transactional and analytic processing illustrating the ability to extract analytic insights from a production database setup – aka “Hybrid Transactional and Analytical Processing (HTAP)”, the holy grail of database analytics.
As the demo shows, the Vexata platform linearly scaled the IOPS and throughput without adding excessive latency to the overall system performance as shown in the VX-Manager analytics screenshot below.
Figure 3: VX-Manager Analytics dashboard with heavy VM workloads
Using 16 ports of 32G Fibre Channel, the VX-100F offers over 50GB/s of read throughput and 25GB/s of write throughput. For these kinds of mixed workload environments, you see that the Vexata VX-100F system delivers bare-metal performance to virtualized applications, easily supporting these scenarios with headroom to allow additional scaling and growth.
A great example of the Vexata solution deployed at a customer environment is based on the Private Cloud deployment at Oath/AOL. Oath Inc. had two tiers of storage for their Oracle and MySQL VMs – one for performance and the other for capacity. The VMs performance would become unpredictable under load and the “noisy neighbor” issue would result in unaligned IOs, and at times the VMs would see huge IO queues, impacting performance and availability. Oath chose Vexata for their private cloud consolidation because the performance at scale ensured that Vexata would maintain line rate performance across their Fibre Channel infrastructure. Vexata was able to consolidate the infrastructure to a single tier and also address any performance issues.
As this post illustrates, the only way to fully utilize VMware deployments is to ensure that the underlying storage infrastructure does not create bottlenecks to deliver the I/O to the virtualized workloads. With that accomplished, architects will find that they have more confidence.
If you are at VMworld this week; hit me up at @RickWalsworth and we can go grab a #vbeer in Vegas together. You can learn more about Vexata solutions for VMware by downloading the VMware solution brief here. Stay tuned for more updates from the VMworld show floor over the coming days.