Here at QLogic there are two sides to my role. Like most CIOs, I am the consumer of IT Solutions. The other side of my role is also to create and communicate I/O solutions for our IT customers. That's to be the theme of 'C I/O perspective' - expressing the value proposition and best practices of IO solutions for IT shops. With that, let me dive into the discussion on why CIOs should care about I/O.
TCO and Service Levels
The CIO's report card is based upon achieving the right balance between low Total Cost of Ownership (TCO) and consistent Service Levels. Hitting the best Service Levels without any credence to cost is not sustainable, but endless cost cuts with little attention to Service Levels could devastate the business.
Traditionally, Enterprise ITs and Service Providers juggled with CPU, memory, and storage as three main variables that optimize the infrastructure. However, with the increasing VM density in a Virtualized environment and the advent of multi-tenancy in a Cloud environment, I/O constraints are taking center stage now. The impact of properly controlling network and storage I/O at various levels of the environment continues to become increasingly important. In fact, I can argue that I/O may be the most important control variable in data centers today, as it controls the entry to the Network – which takes the lion's share of the data center infrastructure and operations cost per Gartner.
Industry perspective on I/O
If you don't believe, listen to the collective voice of the End Users in Open Data Center Alliance (ODCA). ODCA is an End User community, comprised of large Enterprises (like Disney and BMW) and Service Providers (like Verizon and TSystem), that pull together requirements on the most burning problems in the Cloud with a goal of resolving those issues in an Open and Multi-Vendor manner. One of the first few Usage Models that they rolled out was on I/O Controls. Lack of granular control, per ODCA, could result in uncontrolled contention between VMs (i.e., a noisy neighbor) and failure to meet quality of service (QoS) targets for an application or workload. To compensate for the fear of lack of QoS, most IT organizations overprovision their infrastructure, which results in higher Total Cost of Ownership or TCO.
How QLogic IT addresses this problem
I would classify ours as a typical mid-size IT. We have about 300 servers (fully virtualized) and about 1000 VMs, which host engineering, office, and business workloads. We were Virtualization early-adopters and reaping reasonable benefits from it. However, the cost of maintaining separate SAN and Network fabric was high. So, we followed the industry best practice of moving to 10G and a converged network. However, our Server utilization continued to be low. We were overprovisioning servers and I/O to maintain the desired Service Levels. Dev / Test environments were maintained on separate Server clusters, so as to not interfere with Production Service Levels, but this again added to inefficiencies. Mission Critical Databases were particularly overprovisioned for the same reason.
Our solution was as such: we adopted hardware-based network partitioning (or NPAR) [available in QLogic FastLinQ 10GbE Adapters]. I will save the details of NPAR for future blogs. NPAR based solution on I/O control helped us in the following ways.
- We increased server consolidation by 20% by provisioning VMs more deterministically based on guaranteed QoS provided by NPAR. We were able to get rid of a separate Dev / Test cluster, further easing the maintenance pain.
- We have achieved predictable database (and other critical applications) performance by provisioning on NPAR based guaranteed I/O partitions. This ensured predictable performance (and thereby Service Levels) without the need for overprovisioning. Our lab test found significant performance degradation when using a Network adapter without NPAR based partitioning. I encourage you to read the complete whitepaper for details.
Hopefully, I have convinced you of the need for better I/O control in your IT shop. It goes a long way in achieving desired service levels, while keeping TCO low. You get the ability to control I/O partitions (NPAR) for free with QLogic Network Adapters. Although our IT implementation was on VMware, you can implement on any Hypervisor, even bare metal. Drop me a comment if you have questions.
Look out for more C I/O perspective in future blogs!