Emulex Blog: Emulex Labs

SDN Killed the Radio Star

Posted June 19th, 2013 by Sonny Singh

The only reason I recall it was 1984 was because Van Halen happened to christen their seminal album the very same name (you guessed it, 1984). And why was this year and album so monumental to a young lad named Sonny? Well it’s because it also happened to mark the first time I was presented with my very own Sony Walkman. You see, the Walkman embodied so much more than a piece of hardware to play music…it represented auditory freedom to which I was able to escape to from the daily high-pressures that came with being a 4th grader (e.g. Super Sugar Crisp or Captain Crunch for breakfast? G.I. Joe or the Smurfs at 3:00 p.m.?).

It’s only befitting to marvel at how far the mechanisms for delivering music have come. Disruptive technology like those provided by the likes of file-sharing revolutionaries Napster in the late 90’s rocked the music industry to the core and what was once unfathomable is now commonplace with software-driven conduits like Spotify, Pandora and other streaming music services like the newly revamped iTunes Radio becoming the norm for delivering your favorite music. What makes the on-demand music streaming market so enticing is you are no longer dependent on having to lug around bulky mediums like compact disks and being confined to all of 13 songs at any given time. With online streaming music, you have access to millions of songs and thousands of cataloged albums and artists for a nominal monthly fee. I can create and access my own playlists based on my many moods: Death metal when taking my kids to their dreaded doctor appointments or some Bach to replenish my depleted brain cells upon the conclusion of my kids’ doctor appointments. I can access music any time I want from my iPhone, work or home computers and even my car.

Forget Penicillin, online music streaming in my humble opinion has proven to be the most impactful event to human civilization in this modern era.

Software Defined Networks (SDN) – Network Switches Run for your Lives!

SDNs are not just another “techy” buzz term being propagated at the latest industry events. In fact, SDNs are very similar to online music streaming in regards to the paradigm shift they will soon cause within their own respective ecosystem.

Quick primer: SDN is basically an approach to networking in which the control plane is decoupled from data plane and given to a software application.. SDN promises to deliver significant benefits including reduced costs and better efficiency to network operations by introducing a layer of software between bare metal network components and the network admins who configure and monitor them. What makes this technology so revolutionary is network admins can now configure and make adjustments to their network devices through a software interface instead of having to manually configure cumbersome hardware and actually physically access the network devices. SDN provides consistent, relatively fast network management by enabling network-wide changes from a single management console.

In today’s current network topologies, proprietary firmware residing on the switch determines where packets of data are sent. With SDN, the network admin can actually play traffic cop and dictate where they want their data to go via a centralized network console that integrates the information and controls all their network switches. With SDN, network admins are no longer at the mercy of expensive switches with proprietary firmware that must be configured manually. They can now configure and set traffic patterns on the fly where and when they want to.

One more thing, the giants of the tech industry have begun to whole-heartedly embrace SDN initiatives including Cisco, IBM, Alactel, Juniper Networks, Broadcom, Citrix Dell, Google, HP, Intel, NEC and Verizon (to name just a few). As an example, VMware recently acquired a SDN based startup called Nicira for a whopping $1.2B to no doubt go after Cisco’s switch market.

Don’t get me wrong; will SDN put switch manufactures out of business? Absolutely not (at least not in the short-term). SDN still have a long way to go before becoming a mature technology with major vendors still having yet to agree on a common set of interoperability standards for all their networked product lines. Until these standards are cemented unilaterally, you will mostly see the early adopters like Google and Amazon drive SDN adoption across their ecosystems. However, you better believe SDN is on every big tech player’s roadmap.

In short, SDN, like online streaming music, has everyone’s attention. The last thing big tech giants want to do is emulate the stodgy, arrogant and inflexible record executives who wrote off and fought online streaming music rather than adopting it. Yes, record companies are “now” coming along but for many it’s a bit too late and the ship has set sail so to speak. Adapt or die, it’s that simple.

Emulex – We understand and embrace SDN

At Emulex, we understand that the I/O system presents a new area of programmability and organizations need to think about how they can leverage I/O systems and architectures to deliver services and improve on end-to-end performance. Organizations need to stay one step ahead of the ever-changing needs of today’s demanding data centers. As you pursue IT initiatives, such as SDN, the experts at Emulex can help by providing the right products, support, service and knowledge transfer to help you fulfill your objectives.

Enterprise Storage Group (ESG) provides excellent information in their Book Project for organizations looking to plan and/or maximize their SDN deployments. Download the full Book Project for the detailed report on Why I/O is strategic to SDN. http://www.emulex.com/solutions/it-initiatives/software-defined-networking.html

Network Virtualization Explanation for the Masses – Scotty, Beam Me Up!

Posted June 14th, 2013 by Mike Jochimsen

Last week Emulex announced our Virtual Network EXceleration (VNeX) technology at Microsoft TechEd.  This new technology will be used for transporting virtual machines (VMs) in a Virtual Network Fabric (VNF) environment.  Sounds pretty futuristic to me – like something from Star Trek!

The basic premise is that server virtualization is good.  Server virtualization without the confines of traditional networking is better.  You see, the advent of the transporter really opened the door for Captain Kirk and the crew of the Enterprise to quickly and safely move between locations without having to cramp themselves into those pesky shuttles.  The ability to dematerialize from the transporter room on the Enterprise and rematerialize on the surface of countless planets with only a 4-5 second delay!  How many times did we see Kirk and Spock rematerialize and continue a conversation they were having just prior to the transporting process?

But I digress.  What the heck does this have to do with VNeX?  With Microsoft and VMware releasing their network virtualization technologies, Network Virtualization using Generic Routing Encapsulation (NVGRE) and Virtual Extensible LAN (VXLAN), the traditional boundaries of a level 2 network have finally been overcome.  These solutions allow a VM to become portable across a level 3 virtual network, regardless of its underlying level 2 network, and regardless of its physical location.  This means that private and hybrid cloud environments, which need freedom to have different physical locations appear to be one virtual location, become practical.  Simply put, a VM in San Francisco can migrate to a server in London without any reconfiguration or new IP address assignment.

There are a lot of virtual networking solutions out there, and Emulex’s focus is how Emulex makes virtual networking better with VNeX.  The process that both NVGRE and VXLAN go through to transport that VM from one physical network to another requires a tremendous amount of processing power.  While Moore’s Law might allow for the doubling of processing power on a chip every 18 months, this process can tax even the latest CPUs.  Our testing has shown that the process to encapsulate these VMs for transportation can add up to a 27% tax on network throughput due to the CPU utilization it requires.

With VNeX, Emulex will be able to recapture the vast majority of that network bandwidth by offloading critical processing from the host CPU to the NIC. For a more thorough explanation of this, I encourage you to read the Emulex Labs blog that Brandon Hoff wrote last week.

So, in effect, VNeX is the Heisenberg Compensator portion of the transporter!  Without it, the Heisenberg uncertainty principal would certainly make for some interesting rematerializations, since the transporter wouldn’t be able to account for the position or momentum of target particles.  Not an attractive visual…

The message is clear – don’t let the Heisenberg uncertainty principal byte you in the network.  Emulex’s Heisenberg Compensator, I mean VNeX, will allow you to avoid those embarrassing moments and ensure you get the right head on the right body.

The Road to Gen 5 Fibre Channel

Posted June 13th, 2013 by Emulex

By: Scott Shimomura, Director of Product Marketing, Data Center Storage and Solutions, Brocade

Fibre Channel has long been defined by speed. Paradoxically, when you ask SAN customers why they continue to buy Fibre Channel, there’s a six-nines probability that they will say it’s because of reliability or availability. At best, speed is a distant second or third choice; it’s a stereotype that has stuck in the progression from 1, 2, 4, 8 and now 16 Gbps Fibre Channel technology.

Gen 5 Fibre Channel is the new name we introduced for 16 Gbps technology. Our intent was to connect with customers and partners through marketing messaging that changes the focus from speed (in this case, 16 Gbps) to the underlying technology (Fibre Channel) and capabilities.

This simple name change has started a movement within the industry to change the narrative and discussions around this important data center technology. Emulex’s introduction of Gen 5 Fibre Channel Host Bus Adapters (HBAs) is an important milestone and signal within the industry. Together, our message is that Fibre Channel is not just about speeds and feeds; it’s about a new class of technology defined by innovation and unique capabilities that solve customer challenges.

High density virtualization requires unprecedented reliability in addition to performance due to the increased impact of downtime and outages. Infrastructure complexity is driving requirements for consolidation and simplification of network resources. Adoption of solid state disk (SSD) and flash storage will outpace the I/O and bandwidth demands of legacy SAN infrastructure, and drive the need for better monitoring and diagnostic tools to ensure optimized performance.

Brocade and Emulex Gen 5 Fibre Channel technology is designed to address these evolving challenges of enterprise storage. It’s a purpose-built and data center-proven network that delivers unmatched reliability, simplicity, and low-latency 16 Gbps performance. Our portfolios of Gen 5 Fibre Channel backbones, switches, and adapters will unleash the full potential of high-density server virtualization, cloud architectures, and next-generation storage.

Follow Brocade and Emulex as we chart a new and exciting path for data center networking technology.

Brocade on Twitter @BRCDcomm

I am on Twitter @scottshimo

Gen 5 Fibre Channel. What’s In A Name?

Posted June 12th, 2013 by Barbara Porter

Emulex has moved to a new generational-based naming scheme for its LightPulse Fibre Channel Host Bus Adapters (HBAs). So why the change? The currently shipping fifth generation (Gen 5) Fibre Channel HBAs pack more than just a 16Gb punch as the previous name leads you to believe.

LightPulse Gen 5 Fibre Channel HBAs have been designed from the ground up for the virtualization, cloud and database era, taking Fibre Channel SANs to the next level by delivering more than just incredible performance and higher throughput. Emulex Gen 5 HBAs deliver:

  • Increased SAN reliability, data integrity and availability
  • Management simplicity
  • Reduced operational costs across the data center
  • Multi-speed adapters
  • Performance acceleration for virtualization and mission critical applications

As you can see, Gen 5 is so much more than incredible performance.  Fibre Channel development has evolved as data center needs have changed. Traditionally storage was measured by Gb/$ metrics, and many technologies have been developed that have helped address the Gb/$ requirement such as data-dedupe technologies, and flash-based storage. Today, data centers are challenged with how to accelerate application performance, maximize investment in costly applications licenses and assure quality of service (QoS) for mission-critical applications, cloud and virtualized deployments. These days, I/O operations per second (IOPS)/$ is a more frequently used metric.

Gen 5 Fibre Channel has addressed the performance acceleration challenge by delivering up to 6x more IOPS, while reducing latency by up to 75%, drastically speeding up access to storage compared to the previous generation 8GFC HBAs.  Labs tests comparing 8GFC and Gen 5 (16GFC) HBA performance have shown that by simply upgrading to Gen 5 HBAs, data centers can get a big application performance boost. Considering the incremental cost to upgrade to Gen 5 HBAs is a few hundred dollars per server, this upgrade provides a simple, cost effective applications performance boost compared to upgrading the entire infrastructure including additional applications licenses.

The chart above shows some of the proven performance advantages.

  • By simply moving to Gen 5 HBAs and switches, you can get up 41% more transactions for your SQL Server applications.  (Download the paper here).
  • Demartek has shown 33% faster completion time for data warehousing workload (download the paper here)
  • VMware ran performance tests comparing 8GFC to Gen 5 (16GFC) and got 75% better throughput (Download the paper here)
  • ESG measured 3X the IOPS for Exchange workloads (Download the paper here)

Leading Fibre Channel vendors Emulex and Brocade  have delivered products that not only run at 16GFC speeds, but also run at 8GFC, 4GFC and 10Gb Ethernet  (10GbE) on the same ASIC. Emulex Gen 5 Fibre Channel encompasses all these speeds and protocols. As you can see, while doubling the FC bandwidth is a fantastic feature built into Gen 5 Fibre Channel, its only part of the story.

The Benefits of Network Virtualization Offload Technologies to Optimize Performance for NVGRE

Posted June 3rd, 2013 by Brandon Hoff

As we have discussed before, NVGRE or Network Virtualization using GRE (an informative RFC) defines how to build virtual networks in Hyper-V environments.  Virtual Network Fabrics (VNF) creates a virtual network infrastructure where a virtual machine (VM) can be created and moved without any limitations that would be imposed by the legacy network infrastructure.

With NVGRE, VMs live on a single virtual network defined by a Tenant Network ID (TNI) in the NVGRE virtual network.  VMs can be moved from any physical server to any physical server and NVGRE creates a virtual L2 network across physical L3 boundaries so that the VM is able to keep its MAC and IP address no matter where it moves.  Furthermore, network configuration becomes automated so that any network changes required to create a new VM can be done in minutes instead of days.  This improves the agility of private and hybrid cloud infrastructures and lowers the costs of network management for private and hybrid cloud environments.

It is important to note that NVGRE can be implemented in software and the solution works well on Converged Network Adapters (CNAs) and Network Interface Cards (NICs) provided by Emulex.  That said, as Microsoft stated in their presentations at the 2011 Windows Build Conference (here and here and slides here) NIC participation in NVGRE, specifically offloads, to build encapsulated packets for NVGRE, is essential.  Today, without NIC participation, GRE breaks today’s task offloads, which disables nearly 10 years of NIC enhancements that improves performance for high performance Ethernet networks.  Specifically, NVGRE breaks LSO and other NIC performance optimizations.  This creates a performance penalty, as illustrated in the table below.

Note:  These test results are illustrative in nature and will vary based on VM density, server configuration, and other test parameters.

Basically, network throughput is reduced by 27% in this test scenario.  This means that VM density may be reduced by 27% which is significant.  Therefore, Emulex and Microsoft have been working together to solve this performance issue.  The solution is to add additional NIC offloads to improve system performance.  Today, Emulex announced our forthcoming Virtual Network Exceleration™ (VNeX)  virtual network offload technology that solves these problems and provides optimal performance for virtual networks.  Emulex virtual network offload technology provides the following:

  1. Send
    1. Checksum computation for IPv4/IPv6 and TCP payload (inner and outer)
    2. LSO V1 and V2
    3. Receive
      1. RSS/VMQ
      2. Checksum validation for IPv4/IPv6 and TCP payload (inner and outer)
      3. And specific Emulex virtual network offload technology integration for NVGRE

An illustration of Emulex’s virtual network offload technology is shown below:

The graph below shows the performance improvement with offloads optimized for NVGRE.

Note:  These test results are illustrative in nature and will vary based on VM density, server configuration, and other test parameters.

Bottom line:  NIC offload for NVGRE significantly improves performance.  NVGRE is a great solution to build virtual network fabrics and Emulex’s virtual network offload technology delivers the following benefits:

  1. Improved VM density for Hyper-V Environments
  2. Better network throughput
  3. Lower CPU utilization when implementing NVGRE

Emulex leads in technologies for virtual network fabrics, and we look forward to providing more proof points to illustrate the benefits of virtual network offload technologies.

Some of these products may not be available in the U.S.  Please contact your supplier for more information.

When is too much bandwidth not enough?

Posted May 31st, 2013 by Mike Jochimsen

There is a lot going on in the backup market these days.  Tape is either dead, dying, or being relegated to an archive tier depending upon who you talk to.  At the same time, backup software sales are growing.

I read a report recently from the Santa Clara Consulting Group, who publishes quarterly and annual reports on tape drive sales, called the Tape Tracker.   According to the 2012 annual report, tape drive sales fell 30% from 2011, on top of the 25% fall between 2010 and 2011.  In fact, 2012 tape drive sales were a mere 40% of tape drive sales in 2007.  The market has been declining fairly steadily (with the exception of an 18% uptick in in 2010).  At the same time, I read an analysis of Quantum’s quarterly earnings, which stated that tape is continuing to impact earnings.

In the same week, I read that Symantec had posted double digit growth in its backup business. Also, Commvault recently reported 23% revenue growth quarter over quarter, 21% revenue growth year-over-year, and projected double digit growth in the coming fiscal year.  On April 24, backup software company Veeam CEO Ratmir Timashev said, “It took Veeam six years to reach 50,000 customers. Less than six months after hitting that milestone, we passed the 60,000 customer mark,” while announcing that it had completed its 27th consecutive quarter of growth.

I think we’ve established that tape and backup software are heading in opposite directions.  Wait, what?  Okay, there is really a simple explanation.

Data volumes continue to grow and the cost of data storage continues to fall.  I told you a few weeks ago in my blog that the adoption of solid state disk (SSD) in the data center as primary storage was increasing at the expense of hard disk drives (HDD). That is because IT is making the paradigm shift from $/GB to $/I/O operations per second (IOPS) as a critical metric in evaluating the true cost of storage.  So, as SSD becomes the preferred repository for tier 1 storage, HDD becomes the perfect repository for backup.  In the same Quantum earnings release, they touted the increase in sales for their disk-based technology.

Coupled with that, backup software is getting smarter about backing up data.  For example, NetBackup Accelerator speeds backup by merging an incremental backup with the last full backup to create a new full backup.  The incremental is done in the host while the merge is done in the media server. This allows the “full” backup to run up to 100x faster than a traditional full backup with NetBackup Accelerator. When you combine Accelerator with NetBackup’s deduplication technology, the performance gets even better.

Symantec has also included virtual machine disk exclusion (VMDK) technology so that VMware backups can now be defined to exclude specific VMDKs from backup processing. Either the boot disk or all data disks can be excluded, which means data that is considered redundant can be skipped. This enables faster backups with decreased backup storage requirements. This doesn’t mean that overall backup volume is decreasing given the increases in data volume.  At best, backup vendors are holding big data at bay temporarily.

It is really two issues which have converged to make IT shops rethink backup strategy.  With IT budgets not recovering from the dark days of 2008, tape drive refresh cycles have gotten much longer.  It makes perfect sense.  If my current tape drive is still working, then I am going to spend my scarce budget on bigger, faster storage.  Combine that with the fact that continued data growth has made the concept of backing up my production data directly to disk during my backup window an ancient (and quaint) idea.  Today, it seems that most IT shops have gone the route of replicating production data to another disk-based storage array, which can then be backed up to tape without impacting production work.  Disk to disk to tape (D2D2T) has been around for years, but it seems like it has become the rule rather than exception as a solution for data sprawl.

In a survey commissioned by Symantec of more than 1400 enterprises around the world, the responses for “what is the primary reason you are not meeting your backup and recovery SLAs” were:

So, you can see the problem – too much data and not enough bandwidth.

Bigger data and faster repositories need bigger I/O pipes to satisfy them.  Emulex 16Gb Fibre Channel (16GFC) Host Bus Adapters (HBAs) provide the increased bandwidth that customers need to feed both the replication (disk to disk) and backup (disk to tape) environments.  The LPe16000B 16GFC HBA provides twice the bandwidth, half the latency and five times the throughput of its predecessor.  This screaming fast performance is ideally suited to this need.

This is why backup companies such as Symantec have certified the Emulex LPe16000B series of Fibre Channel adapter for use with Backup Exec and NetBackup.   Emulex 16GFC offerings are becoming pervasive on servers and Emulex is clearly the leader in 16GFC adapters.  As 16GFC storage targets begin to roll out, end-to-end 16GFC offerings will enable enterprises to meet the demanding needs for their production-replication-backup environments.  We saw the first announcement just last week with the Infortrend EonStor DS G7i Series arrays .  This is just the beginning folks…

Why Flash Storage is Here to Stay – Unlike Fisker or DeLorean

Posted May 24th, 2013 by Sonny Singh

I’ll never forget the first time I laid my eyes on the gorgeous Fisker Karma. Up until that point, the public was relegated to sterile hybrid, electric plug-in automobiles like the Toyota Prius or Nissan Leaf which had about as much sex appeal as Martha Stewart gracing the cover of the next Victoria’s Secret catalog. The Fisker Karma promised the aesthetics and performance of a luxurious automobile out of Modena built upon green technology that would help lessen the impact to our environment. Think of it this way, now Leonardo DiCaprio could invest in a company whose automobile would look just as attractive as the supermodel he brought to the last red carpet event and all the while, maintaining the good graces of his indigenous tribe buddies in the Amazon rainforest. The world was smitten as were private investors and our own government who poured in $171 million of taxpayer money and about $1.1 billion of venture capital (VC) cash.

Unfortunately, Fisker is heading down the same path as the Delorean (the beloved icon of the 80’s), so don’t be surprised if you see it make a cameo in Back to the Future 5. Fisker should be announcing bankruptcy any day considering it still owes the government around $192 million in loans and hasn’t made a car in more than a year. All told, Fisker attracted $1.1 billion in private investment, the vast majority of which took place after it got the first Department of Energy loan.

So why did Fisker fail? It depends who you ask and I want to avoid a political discourse, so let’s stick to the most tangible explanations¹:

  • Fisker’s claimed the Karma would get 67.2 MPG and 50 miles of electric range. If fell significantly short of that claim.
  • The money spent by Fisker compared to the amount of cars produced was approximately $660,000 per vehicle
  • The production goal for 2011 was 7,000…Fisker only shipped 1,500
  • The sole supplier of lithium-ion batteries for the Karma went bankrupt
  • Fisker received one of Consumer Reports’ worst ratings ever for a passenger car
  • Fisker recalled the first 239 Karma plug-in hybrids there were numerous incidents of them catching on fire
  • During a Consumer Reports test drive, the Fisker Karma broke down
  • During Hurricane Sandy, Fisker lost about 300 Karmas

The final nail in the coffin for Fisker was when Justin Bieber reportedly gave his Fisker Karma to a friend while he was on tour because he was sick of it breaking down when trying to run from the paparazzi.

Flash Storage – It is sexy, eco-friendly and here to stay

Being the pseudo tree hugger that I am, I haven’t given up on green technology quite yet, especially in the IT industry. Flash storage is a perfect example of disruptive technology that is innovative, efficient and eco-friendly. Flash storage should be familiar to anyone who has ever used a USB thumb drive and uses electricity, has no mechanical parts and typically only consumes 20% of the power and reads more than 100 times faster than traditional mechanical hard drives (check out this excellent blog by Mike Jochimsen on why solid state disks (SSDs) are the wave of the future). Furthermore, data center managers looking for ways to address the energy drain represented by hard drives are implementing flash storage in their data centers as a way to achieve green computing or green data center benchmarks.

Businesses with I/O-intensive applications have found flash storage to be efficient and cost-effective and that’s why enterprise storage providers like EMC, chip makers like Samsung and server manufacturers like Oracle have all entered the flash storage market. Like the Fisker Karma that retailed at more than $100K, the higher cost of flash technology compared to traditional disk storage warranted a price premium  but more favorable flash economics are spurring broader adoption (unlike the Fisker Karma).  The good news is the price of SSDs falling so flash storage is becomes more attractive by the day and much more accessible to not just the enterprise, but also small-to-medium sized business-level IT organizations.

Key benefits of flash technology²:

  • Flash based storage is faster than hard drives because they handle storage chores electronically, while their traditional counterparts rely on moving parts — specifically a spindle motor that rotates disks and a mechanical arm that reads and writes the data.
  • Flash’s lack of moving parts delivers another benefit: lower energy consumption which equates to better environments footprints.
  • A flash drive-equipped array that provides a 10-speed performance boost is more cost-effective than conventional technology on a cost per input/output operations per second.
  • Aggregating flash as a shared resource on a storage-area network also makes the technology more affordable.

Like any mature technology, the cost of flash memory is coming down, and since storage requirements are always increasing, it isn’t surprising more and more organizations are using flash to store, transport, carry or access data quickly and reliably, whether on the move, in the office or at home. At the same time, flash storage appeals to the environmentally conscious IT organization looking to reduce their impact to the environment.

Emulex – we understand flash technology³

At Emulex, we are beginning to see the adoption of all flash appliances in the market increase as the waves of large databases and increasingly dense virtual environments demand larger data stores with higher I/O operations per second (IOPS) and lower latency.  Organizations including GreenBytes with their IO Offload Engine providing an accelerated virtual desktop infrastructure (VDI), and Fusion-io’s ION Software Defined Storage, which can either present itself as fabric-based cache or primary storage are perfect examples.  On the extreme end of the scale, companies like PureStorage and Violin Memory provide all flash arrays capable of handling very robust and demanding application environments.

Emulex is right in the middle of this evolution.  Our ever-evolving Fibre Channel and Ethernet I/O connectivity solutions continue to provide the high IOPS and low latency that customers need to keep pace with the abilities of today’s high end servers and storage.  Also, as SSDs, flash and cache continue to appear at new points in the I/O path, Emulex is on the leading edge of understanding how to optimize I/O architectures and provide the solutions that will keep pace with the endpoints.

Our world class monitoring solutions, OneCommand Vision and EndaceVision can ensure that the I/O channel is performing optimally and securely.  The combination of these two products gives us this ability across both Fibre Channel and Ethernet fabrics. We continue to evolve these solutions in conjunction with our ecosystem partners to ensure that as databases, hypervisors, big data analytics and cloud computing evolve with technologies such as SSD and flash memory, we can continue to ensure this optimal performance.

Honey…I just bought a used Fisker Karma!

Like most luxurious cars I have lusted after, I’ve usually had to settle for the version made by Hot Wheels or Hasbro. However, things may change. As of this writing, there were about 28 separate ads for Fisker Karmas on eBay alone. It gets even better…the once $110,000 Karma has dropped in most cases by more than half and the lowest bids I’ve come across on eBay start at $40,000, with most purchases ending up between $60-$70K. Sure it wont have a warranty, my wife would leave me and I cant put three baby seats in it, but it would look great in my driveway (car exploding and catching on fire in my driveway be damned).

I want one.


1.  The Washington Examiner,47 depressing facts about Fisker’s epic electric car failure, April 2013

2.  FCW,Storage optimization: Flash finds some government niches,May 2013

3. Emulex Labs blog, May 2013

Happy Birthday, Ethernet!

Posted May 22nd, 2013 by Emulex

Today’s Labs blog comes from Joe Gervais, senior director of product management. He brings us his thoughts on the 40th anniversary of Ethernet today, and draws from his own experience using Ethernet for the past two decades. Take it away, Joe!

~~

Internet cablesToday, at the Xerox Palo Alto Research Center (PARC), the networking industry is celebrating the 40th anniversary of Bob Metcalfe’s invention of Ethernet. In the early 1980s, when Ethernet was first commercialized, it ran at 10 million bits per second. In comparison, the emerging 40Gb Ethernet (40GbE) standard can send an entire frame on the wire in the time the first byte of a frame could be sent with the original Ethernet.

The fascinating part of Ethernet is how it’s fulfilled Metcalfe’s Law – the value of a network is proportional to the square of the number of connected users of the system. In the early days, you might have a department network with 20 users sharing computing resources or files. These Local Area Networks (LANs) would then be connected and become more valuable, and services like email would crop up. Then we saw widespread use of the Internet, and the World Wide Web.

I look at the way my home network is used now versus 20 years ago. In the 1990s, it was connecting my computer to my children’s computers so we could share a dial-up Internet connection or a printer. Today, it’s using my iPhone over Wi-Fi to my Wi-Fi router as the remote for my Ethernet connected stereo and Blu-ray player, with my Internet connection being a fiber optic passive optical network connection. It’s watching a Netflix movie on the television, then pausing the show and changing over to a tablet elsewhere in the house to finish watching the show. It’s my wife, a technophobe, using FaceTime to videoconference with her granddaughter across the country and using Facebook to keep up with far-flung friends.

A number of years ago, Sun Microsystems had a slogan: the Network Is the Computer. In today’s world, the network is central to many aspects of life. Over the past several years I’ve been working with a non-profit supporting Bible translation for the 2000 languages yet to receive a word of the Bible in their own language.  One of the non-profit’s technology initiatives is supplying translation acceleration kits. This is a battery and solar panel, to power a netbook that is connected over Ethernet to a satellite terminal. This allows the translation team to communicate instantly with their translation consultants, saving days of treacherous travel and speeding the translation task by years – all through the use of Ethernet and the Internet.

What does all of this have to do with business computing? Much of this home networking is driving the build-out of massively scalable data centers – services like Amazon’s Web Services (AWS) that provide the infrastructure used by Netflix to stream content to consumers. Google and Microsoft have some of the world’s largest data centers around the planet. This segment is consuming a new class of multinode rack servers that grew nearly 100% between 2011 and 2012 and already make up about 15% of total server shipments1.  Much of this infrastructure is using 10GbE today for server connectivity, and is the fastest interconnect available for Internet connectivity, with 100G transceiver shipments tripling in 20122. This is pushing the Institute of Electrical and Electronics Engineers (IEEE) 802.3 working group to begin work on the next group of standards for Ethernet beyond 100GbE. Projections indicate that today’s social media and mobile devices will drive the need for ten terabit Ethernet by the end of the decade3.

What does the future hold for Ethernet? We’ve seen Ethernet evolve from the early days of being a party line – half duplex using CSMA/CD – carrier sense, multiple access with collision detection (where an Ethernet endpoint attempts to send a packet and if it collides with another packet, backs off and tries again) – to today’s full duplex, point-to-point switched topology. We’ve seen the evolution from coaxial backbones to twisted pair. It’s likely the future holds more mainstream usage of optical interconnect as speeds increase. The optics vendors today are building high volume, low power, lower cost embedded optical solutions called active optical cables that provide optical transceivers connected with a pre-terminated optical cable. Silicon photonics holds promise for optical solutions at 100GbE and faster, where a fixed light is injected into a silicon-based optical switch.  Honestly, I’ve been following Ethernet as a user and producer for nearly 30 years, and it’s hard to project what the next networking technology will look like, but it will most certainly be named Ethernet. Happy Birthday, Ethernet.

  1. Gartner:  Forecast: Servers by Form Factor, Worldwide, 1Q13 Update,  April 11, 2013.
  2. Infonetics:  Optical transceiver market bolstered as 100G arrives in force (press release)
  3. IEEE: IEEE Launches Study Group to Explore 400Gb/s Ethernet (press release)

How Big Data Use is Helping Catch Criminals

Posted May 14th, 2013 by Sonny Singh

In the aftermath of the recent Boston Marathon terrorist attacks, I stumbled across a very interesting article by FCW which provides insight into the latest technology and IT trends being deployed by government agencies. In prior blogs, I’ve mentioned on more than one occasion how the use of “big data” seems to be proliferating more than ever. Big data is defined in simple terms as a means to gather insights from large amounts of data sets and then disseminating those insights into strategic and tactical courses of action. It is actually not surprising that big data practices are being used to help solve crimes simply because in today’s modern age of counter-terrorism, what goes on “behind the firewall” is almost as crucial as what happens at the actual crime scene.

Ultimately, the FBI investigation provided the public a glimpse of how the deployment of big data and data analytics practices is just scratching the surface towards large scale use down the road. Here is a recap of the key takeaways from the article1

  • Less than 24 hours after the two explosions killed three people and injured dozens more at the April 15 Boston Marathon, the FBI had compiled 10 terabytes (TB) of data in hopes of finding needles in haystacks of information that might lead to the suspects.
  • The FBI-led investigation analyzed mountains of cell phone tower call logs, text messages, social media data, photographs and video surveillance footage to quickly pinpoint the suspects.
  • Facial recognition software was being used to compare faces in photographs and video against visa, passport, driver’s license and other databases.
  • While the 10TB of data gathered by investigators seems like a drop in the bucket (the Feds usually work with Petabytes of data), the investigation still presented officials with a large amount of data crunch due to the sheer volume, various types of media and overall complexity of information they were dealing with and requiring a tight window period to analyze.

Dealing with multiple terabytes or more of video, digital images, text message and cell phone records is complex enough as it is. Just imagine how much more of a quagmire is created when you bring social media into the fray? What I found most interesting about this article was that investigators utilized the services of a company called Topsy Labs to sift through billions of tweets. Topsy has stored every tweet generated since July of 2010, and in the case of this terrorist investigation, allowed investigators to run big-data analytics of Boston-related tweets against hundreds of billions of past and present messages. Topsy’s database analytic software allowed investigators to search every reference ever made to Twitter of the word “bomb” in a specific region including Boston and its adjoining suburbs.

Ultimately, this type of detailed search turned up deleted bomb references from both suspects’ Twitter accounts. This type of search through public records likely revealed additional clues that proved detrimental to the investigation, including which users re-tweeted the bomb mentions or engaged in incriminating dialogue with the terrorist suspects. Furthermore, Topsy has “geo-inferencing” technology which allowed the investigators to accurately map where specified tweets originated (pretty cool considering only about 1% of Twitter users geo-tag their tweets). According to Topsy, those capabilities make it 20 times more accurate than standard Twitter location data.

How amazing is that?

Emulex – We ‘get’ big data

At Emulex, we believe the heart of big data at the core, lies within the framework of an organizations’ network. There are thousands of servers performing parallel processing to create value and those servers talk to each other over Ethernet and Fibre Channel protocols. As such, the latency and throughput of the network’s traffic is the critical path for fast results in big data deployments. Emulex solves these latency issues and is the chosen vendor by organizations worldwide because we provide the right I/O solution to maximize data clusters and allow for the seamless deployment of big data solutions. For a more in-depth view on Emulex’s big data expertise please reference my earlier blog here.

It is unfortunate that we live in a world of uncertainty, fear and carnage at the behest of a few loathsome individuals. But it’s also refreshing to remember our community is capable of greatness and benevolence in times of need just as the citizens of Boston demonstrated during and after this terrible incident. Even though the deployment of big data practices were crucial towards the investigation, let’s not forget it was ultimately the tip of a citizen that finally led investigators to the two perpetrators. In the end, no technology, no matter how advanced, can replace the fortitude and good-will of mankind.

¹ FCW, APR, 2013

Is it time for SSD in the data center? You bet your OPEX!

Posted May 3rd, 2013 by Mike Jochimsen

Recently, we published a slideshow on IT Business Edge titled “Five reasons why HDD is dead and SSD is taking over.”  Provocative?  Sure, but that was the point. Do I really think the hard disk drive (HDD) market is dead?  Not that it matters what I think, but no… EMC, IBM, HP and a huge number of storage vendors continue to sell massive quantities of HDDs every day and will continue doing so for the foreseeable future.

However, recently it feels like we are rapidly approaching Gladwell’s tipping point where “ideas and products and messages and behaviors spread like viruses do1.” Well, the pandemic that is solid state disk (SSD) sure seems to fit that criteria.  Come on, would IBM bet one BILLION dollars in something that is just a fad?

In the immortal words of Ron Popeil, “but wait, there’s more,” it seems like every analyst on the planet is now talking as if the use of SSD/flash in servers and storage is becoming a de facto standard.

So, what is the use case?  While SSDs began appearing in servers in recent years as local storage, the idea that they could effectively replace storage area networks (SANs) began to fade when users realized that large databases, virtual environments, and big data analytics required lots of servers touching common shared storage.  The use case for this local flash storage morphed into server-based caching, which is how companies such as EMC and Fusion-io are now positioning their PCI Express (PCIe)-based flash adapters. Also in recent years, these SSDs began appearing in the storage fabric in at least three use cases – SSD front-ending traditional spinning disks in a storage enclosure (hybrid arrays), SSDs in a fabric-based appliance front-ending a traditional spinning disk array, and all flash arrays as primary storage.

Most of the large storage vendors already offer hybrid arrays, which offer a flash front-end to spinning disks to accelerate performance.  This offers a relatively easy way to begin experiencing the benefits of SSD.  Although this wasn’t without some angst since the addition of SSD into the array created need for changes to the RAID controllers typically optimized to place data on spinning disks.  While some customers see this as an onboarding opportunity into SSD storage, others may find it is their destination, giving “good enough” performance.

We are beginning to see the adoption of all flash appliances in the market increase as the waves of large databases and increasingly dense virtual environments demand larger data stores with higher I/O operations per second (IOPS) and lower latency.  Offerings include GreenBytes with their IO Offload Engine providing an accelerated virtual desktop infrastructure (VDI), and Fusion-io’s ION Software Defined Storage, which can either present itself as fabric-based cache or primary storage.  On the extreme end of the scale, companies like PureStorage and Violin Memory provide all flash arrays capable of handling very robust and demanding application environments.

Again, in the immortal words of Clara Peller “Where’s the beef?”

Much of the resistance to bringing SSD into primary storage has been cost.  Generally, HDDs are ten times more cost effective than SSD when measured on a dollar per GB ($/GB) basis (although Steve Mills of IBM claimed the cost differential was less than double in the article mentioned above. Since SSD costs continue to fall, assume the real answer lies somewhere between 2X and 10X.)  However, SSDs are at least 100 times more cost-effective than HDDs on a dollar per IOPS ($/IOPS) basis. Given that increased IOPS is such a central value proposition to SSD storage, it makes sense for SSD array vendors to make this argument.  But it also makes sense in general.  In the HDD world, generations of spinning disk offered modest improvements in performance when compared to the dramatic leap offered by the move from HDD to SSD.  For example, from 7K, to 10K, to 15K, HDD IOPS increased in modest increments, like 30-100%.  However, the difference in IOPS between a 15K drive and an SSD can be in the range of 2000%!

The other major advantages that SSDs bring to the data center are power and cooling efficiency.  An SSD has no moving parts so it requires much less power to run and generates little to no heat, which has a dramatic effect on cooling costs.  In our slide show referenced above, we quote Marc Staimer, founder of Dragon Slayer Consulting, who estimates the power and cooling savings of SSD over HDD can be in the 50-80% range.  I’ve seen similar estimates from other analysts as well, giving the claims of the flash/SSD community some credence.  This becomes a very important factor given the constraints that today’s data centers are facing in power and cooling capacity.  Given the overall economic situation in recent years, there has been a slowdown in data center build out, forcing many companies to optimize the use of their current facilities.  This provides one more arrow in their quiver of optimization strategies.

So, is HDD technology really dead?  Not likely in the near future.  However, with the growing capacity, shrinking cost and efficiency benefits that SSD can bring to the data center, it is quite conceivable that the tipping point HAS been reached and the mix of HDD vs. SSD in the data center will soon start to fall in favor of SSD technology.  Enterprise customers do need the operational expenditure (OPEX) savings and facilities relief that SSD technology can bring to their “shrink IT while maintaining service level agreement (SLA)” directives.

Emulex is right in the middle of this evolution.  Our ever-evolving Fibre Channel and Ethernet I/O connectivity solutions continue to provide the high IOPS and low latency that customers need to keep pace with the abilities of today’s high end servers and storage.  Also, as SSDs, flash and cache continue to appear at new points in the I/O path, Emulex is on the leading edge of understanding how to optimize I/O architectures and provide the solutions that will keep pace with the endpoints.

Our world class monitoring solutions, OneCommand Vision and EndaceVision can ensure that the I/O channel is performing optimally and securely.  The combination of these two products gives us this ability across both Fibre Channel and Ethernet fabrics. We continue to evolve these solutions in conjunction with our ecosystem partners to ensure that as databases, hypervisors, big data analytics and cloud computing evolve with technologies such as SSD and flash memory, we can continue to ensure this optimal performance.

If you are planning to attend Interop May 7th through 9th in Las Vegas, stop by the Emulex booth #1759 and hear more about what we are doing (or read this blog on what we’re highlighting!).

  1. Malcolm Gladwell “The Tipping Point”, 2000, Little Brown
«Older Posts