Emulex Blog: Emulex Labs

Emulex Labs Discusses Common I/O Performance Issues

Posted December 10th, 2009 by Emulex Labs Team

IT organizations worldwide are challenged to keep their Storage Area Networks (SANs) always available and fully optimized. There is a running joke within the data center community that whenever there is a performance problem, blame the storage network first. There is a reason for this. As the industry transitioned from Direct Attached Storage to Network-Attached Storage and SANs, the storage environment became more complex. There were more components, more vendors, more integration, and as a result, the storage network did introduce more performance problems. Having said that, the fact still remains that not all problems are storage network-related! Often, a fair amount of time can be spent on proving the storage network is, in fact, healthy. In such cases, it is only after this step that root cause analysis can start elsewhere, in earnest.

When customers have performance issues within their data centers, Emulex is often asked to help determine the root cause. This is primarily because of Emulex’s decade-long experience in server and storage network connectivity. Even though the adapter is a key component in the SAN data path, most often, it is not the cause of the problem. Often, the source of the problem resides within the various layers of the system. Layers include the application, files system, protocol stacks, operating systems, schedulers, hypervisors, volume management software, multi-pathing software, device drivers, PCI subsystems, virtual and physical networks, storage devices and more. These layers can all cause various types of issues within the data center, and below, we’ve highlighted some common ones:
Continue reading…

Benchmarking is Our Obsession. Ixia Helps Write the Black Book on CNAs.

Posted October 28th, 2009 by Emulex Labs Team

At Emulex Labs, benchmarking is our obsession. We are experts on benchmarking Fibre Channel. We run Iometer in our sleep. With the launch of our OneConnect™ Universal Converged Network Adapter (UCNA), we have become obsessed with IP benchmarking.

After examining some freeware tools, we found that the finest solution appeared to be from Ixia. Ixia’s highly scalable solutions generate, capture, characterize and emulate network and application traffic, establishing definitive performance and conformance metrics of network devices or systems under test.

We focused in on IxChariot, which contains Application Scripts to measure end-to-end network performance, the response time and throughput of application transactions across the network.

Ixia provides a library of free, pre-defined scripts that you can easily, extensively customize to reflect the unique character of application traffic on your network. Application Scanner lets you automatically create scripts for any custom or commercial application without having to understand the scripting language.

IxChariot:

  • Creates Transmission Control Protocol (TCP), User Diagram Protocol (UDP) and RTP IxChariot scripts and Ixia hardware stream definition files from line traces generated by protocol analyzers
  • Creates scripts for TCP and UDP (including RTP) applications that use Winsock calls
  • Creates IxChariot application group definitions from Winsock traces and from line traces generated by protocol analyzers
  • Automatically generates scripts based on a Winsock trace, performing optimization where needed
  • Traces all Winsock traffic to help you better understand the interaction of applications with multiple connections or processes
  • Enables generation of IxChariot scripts that preserve the timing delays contained in the source trace files.
  • Enables creation of scripts for applications running on non-Windows clients
  • Presents details about the traces you record
  • Enables you to select a single application stream from a line trace, create a script for that stream and edit the script as required
  • Allows you to combine multiple streams into a single script
  • Loads line trace files that contain IPv6 network traffic, and generates Ixia stream definitions corresponding to the IPv6 traffic
  • Generates IxChariot scripts that include large payload files

Ixia’s customer support was phenomenal. As we spoke, they became quite interested in our benchmarking Fibre Channel over Ethernet (FCoE). Our discussions led to joint development in the area of benchmarking processes for the new converged networks. Together, we drafted an Ixia Black Book on CNAs.

As a primer on the technology and testing of CNAs, the Black Book begins with an introduction to CNA technology, the developments that led to the FCoE protocol and data center network convergence. It includes detailed test plans using Ixia’s IxVM, a tool for measuring the performance of virtualized systems, which guides the reader through key tests required to thoroughly measure the performance of a CNA. Step-by-step instructions include using IxChariot™, a component of IxVM, to measure TCP throughput, UDP throughput, latency and storage input/output (I/O) performance in virtualized and non-virtualized environments.

If you are interested in benchmarking converged networks, download a free copy.

To learn more about Ixia Black Books, visit: blackbook.ixiacom.com/blackbook.html

More than Meets the Eye: Performance Numbers for Virtual Server and Transactional Environments

Posted September 25th, 2009 by Emulex Labs Team

Too often in our industry, performance numbers are thrown around with little thought about what they actually mean to someone using the Host Bus Adapters (HBAs) on a day-to-day basis. Recently, Brocade published dazzling results for one of their 8Gb/s HBAs. Dig a little deeper, however, and one finds that those numbers do nothing for someone running a Storage Area Network (SAN) under real-world conditions.

The recently released Brocade numbers are based on a measurement tool called an IOMeter. Based on that performance, they claimed performance leadership in just about any application. While IOMeter is a valid tool, IOPS performance is not a representative measure of performance in virtual server and transactional environments. Here is why Brocade’s methodology is flawed:
Continue reading…

New Emulex vSphere 4 Driver Delivers Near Bare-metal Performance

Posted September 2nd, 2009 by Emulex Labs Team

sanVirtBannerrBlogVMware is hosting VMworld this week in San Francisco. At VMworld, Emulex, Cisco and VMware are launching the SAN Virtuosity series, which will feature white papers and Webcasts that will help turn you into a storage area network (SAN) virtuoso.

In parallel, Emulex is also demonstrating a new vSphere 4 driver for Emulex LightPulse® Fibre Channel host bus adapters (HBAs) and converged network adapters (CNAs), which combine enhancements to the VMware vSphere 4 storage architecture with new Emulex-developed optimizations to provide dramatic improvements in I/O performance.
Continue reading…

Replace Your Server Clunkers

Posted August 19th, 2009 by Emulex Labs Team

“Cash for Clunkers” is arguably one of the hottest government-funded programs in history. Through this program, old, energy-inefficient gas guzzlers are being replaced by newer, greener vehicles.

Like the “Cash for Clunkers” program, the new Intel® Xeon® processor 5500 series (codenamed Nehalem-EP) enables a similar benefit for IT. Replacing older Intel servers with new servers using the Intel Xeon processor 5500 series will allow IT organizations to realize an estimated 90% reduction in annual energy OpEx, letting them recoup their CapEx within as little as eight months.

In addition to energy efficiency, the new Xeon processor 5500 systems have a significant increase in performance due to an integrated memory controller with dedicated memory and links to other processors and the I/O hub leveraging high-speed Intel QuickPath Interface (QPI) connections. When combined with other features such as PCI Express 2.0, Intel Hyper-Threading Technology and Virtualization Technology for Directed I/O (VT-d), the result is a quantum improvement in performance.

When you combine these high-performance systems with new 8 Gb/s Fibre Channel storage area networks (SANs), you can build a platform with unparalleled price/performance.

Emulex Labs recently conducted some performance analysis, looking at performance on an Intel Xeon 5500 versus that of last year’s Intel Xeon 5300 using the Emulex LPe12000 8 Gb/s Fibre Channel Host Bus Adapter (HBA). We also compared the performance to similar products from QLogic.

Highlights include:

  • A 225% increase in IOPs per GHz of CPU resource for Intel Xeon processor 5500 series with Emulex LPe12000 8Gb/s Fibre Channel HBAs
  • 3.4 times the maximum IOPs for Intel Xeon processor 5500 series with Emulex LPe12000 8Gb/s Fibre Channel HBAs
  • 21% higher read IOPs and 29% higher write IOPs for Emulex LPe12002 8Gb/s Fibre Channel HBAs when compared with QLogic’s QLE2562 8Gb/s

For the performance report, click here to download:  Performance Advantages with LPe12000 and Intel Xeon 5500.

For more information on how Intel Xeon 5500 Series Platforms and Emulex 8 Gb/s HBAs perform together, visit Emulex’s Intel Xeon 5500 Series Platforms Solution site.

Sometimes, Being Hot Isn't So Cool

Posted August 6th, 2009 by Emulex Labs Team

For the last several months, there has been a lively debate in the industry about Emulex’s choice to use heat sinks on our 8Gb/s host bus adapters (HBAs). In contrast, QLogic, another market participant, chose not to put a passive cooling device on their HBA. But their decision can leave customers burned.

In today’s mission-critical data centers, customers demand the highest levels of reliability without compromise. But our competition has chosen to take shortcuts with their product that can have IT administrators seeing red, especially when it puts their environments at risk. At Emulex, we feel strongly that keeping our adapters cool in a broad range of server environments is critical to data center success. This is why we used heat sinks when we designed our family of 8Gb/s Fibre Channel HBAs.

Heat sinks are not new to server environments. They are extensively used in blade, pedestal and rack mount servers to ensure proper operation. Heat sinks are applied on a server’s central processing unit (CPU), memory modules, power components and even networking devices (NIC). Since heat sinks are passive, with no moving parts, their inherent design has no negative impact on server reliability. In fact, their use can be considered as an additional safeguard to help improve server reliability. Vendors who properly understand the benefits of heat sinks leverage them to help maximize server performance and reliability.

To help visualize the dramatic difference between our implementation and that of the competition, Emulex Labs used a thermal camera to view both Emulex and QLogic’s HBAs. Our findings were that the Emulex LPe12002 HBA operates 27° F cooler than the QLogic QLE2562 HBA. The image on the below right shows the thermal “hot spots” on the QLogic QLE2562 HBA where higher temperatures were detected.

We even went a step further, removing the heat sink to demonstrate that the Emulex LPe12002 operated cooler than the competition. The images below clearly show that, even without a heat sink, the Emulex LPe12002’s I/O processor operated 10% cooler than QLogic’s.

You can download a flier that outlines the results of our evaluation:
Click to download

If you want to know more about the use of heat sinks, we invite you to watch this video of Charlie Nogalas, vice president of engineering at Emulex, who answers common questions on the use of heat sinks and the performance of our 8Gb/s HBAs.

Heat Sink Video with Charlie Nogales

Additionally , you can access a presentation that shows more detail on our thermal evaluations:
Click to download

Finally, we have provided an FAQ that will answer questions you might have on passive cooling and Emulex HBAs:
Click to download

At Emulex, we are focused on delivering high-quality solutions to our customers and partners without  compromising performance or reliability. This means no short cuts!   We closely scrutinize our product designs to ensure trouble free operation in real-world environments. By deploying heat sinks in our solutions, we are helping customers to improve data center reliability  and keep their businesses running at full speed.  With our 8Gb/s HBAs, you will not find a burning hole in your server, or in your wallet.

Welcome to Emulex Labs

Posted July 24th, 2009 by Emulex Labs Team

Emulex Labs is not your ordinary corporate blog. Emulex Labs is about bits instead of wits, and specs over spin.

Emulex Labs includes the collective of Emulex engineers and techies who benchmark, configure, test, demonstrate and support Emulex adapters for a living. They also explore the functionality, performance and potential limitations of our competitors’ product offerings. They are the “Real Gurus” of connectivity.

This blog is a forum for sharing our tribal knowledge on network convergence, Fibre Channel, Ethernet, management and configuration of Emulex product capabilities, and G2 on competitive products. We will share information on currently shipping products, as well as some that are still in the labs. For IT folks, we will share helpful hints on how to make your storage area network (SAN) run better or provide undocumented time-saving tips. Our labs are overflowing with relevant, yet not widely communicated, stuff that you will hopefully find interesting. If you have your own findings or information, we invite you to share them with us. Or, if you have a question, e-mail it to emulex-labs@emulex.com.

This week, Emulex is hunkering down, preparing for the release of our OneConnect Universal Converged Network Adapter (UCNA).

Thao Vo, one of our curious Technical Marketing Managers, was eager to measure the relative performance of our new UCNA compared to our 8Gb/s and 4Gb/s host bus adapters (HBAs) in terms of IOPS and bandwidth, so he plugged an Emulex LPe11000 4Gb/s Fibre Channel HBA, Emulex LPe12000 8Gb/s Fibre Channel HBAs and Emulex OneConnect 10Gb/s Universal CNA to Cisco Nexus 5000 Series Switch and configured it, using failover multipathing, to a common set of targets. Thao used IOmeter to benchmark the performance and created this video.

Watch it. We think you will be surprised.