Inside Enterprise IT

Virtualization has helped improve hardware utilization tremendously, but its full potential has not yet been realized. Virtualization technologies all share and virtualize a single physical port of the network adapter through software for the I/O needs of the virtual machines. All of that software battles to make I/O decisions for the virtual machines, which causes traffic jams, which slow the I/O performance and limit the number of virtual machines a physical server can run.

To address this issue, PCI-SIG, the special interest group that owns and manages PCI specifications as open industry standards, introduced a suite of specifications for Single Root I/O Virtualization (SR-IOV) specifications to allow multiple operating systems to share a physical interconnect.

Now Dell, Intel and Citrix are putting the PCI-SIG specs into action and will be demonstrating SR-IOV technology at the Intel Developers Forum 2009. IDF attendees can check out the showcase at exhibit booth #711 and hear all the technical brilliance straight from Dell, Intel and Citrix experts.

Here is the gist of how it works. SR-IOV allows hardware implementation of virtual network interface cards (NICs) or virtual functions without software emulation. In this way, a single I/O hardware is subdivided logically to appear as up to 256 virtual NICs and each virtual function is assigned independently and directly to a virtual machine, bypassing the software bottlenecks in the hypervisor to achieve near native performance. It also provides precise per-VM control for the connection speed and QoS. This specific demo uses Intel VT-d and Intel® 82599 10 Gigabit Ethernet device and allows multiple virtual machines (VMs) running on Citrix XenServer on Dell PowerEdge R710 with Intel® Xeon® processor 5500 to directly share I/O devices.

SR-IOV holds great promise. It improves data throughput and increase performance, efficiency, and scalability with high performance I/O devices, while preserving flexibility and mobility with live migration support. In preliminary lab tests with Dell, Intel and Citrix, this technology reduces processor utilization, Increases bandwidth to virtual machines and improves data transfer rates across the network.

Another benefit is that SR-IOV can save reduce capital expenditure as it doesn’t require special network infrastructure. It works with existing standard 10GE Ethernet switches on existing network infrastructures. Some other virtualization solutions rely on proprietary technologies requiring rip-and-replace upgrades to networks to get the benefits of I/O virtualization. SR-IOV technology further cuts hardware costs by reducing the number of physical network cards, and switch ports, as well as cabling for even more consolidation in a virtualized environment.

Rich Hernandez, PG Development Engineer Strategist in Dell Server Advanced Engineering Group has been working on this technology. Here he is with a brief overview:

A Dell employee emailed from the Microsoft Management Summit this morning:

I am in the opening keynote here at MMS. Being the featured hardware partner for SCCM is awesome. The room is PACKED and attendees are hearing Microsoft talk about how great Dell systems management is.

Microsoft’s Bob Muglia, MS SVP, highlighted Dell’s preliminary bare metal deployment pack in his keynote. Hubba hubba and hey now.

Many of our customers use Microsoft’s System Center and Dell’s OpenManage to manage their infrastructure. And who do you think offers the most comprehensive device management through Microsoft’s System Center Suite? (Dell does) And we’re not resting. This week we’re announcing the upcoming availability of the latest Dell Management Packs for Microsoft System Center Operations Manager that will give customers superior monitoring and control of Dell desktops, portables, workstations, rack and tower servers, blades, networked storage devices and even printers.

Our new plug-ins and our roadmap to integrate Dell Services with Microsoft products means our mutual customers will continue to have industry leading tools to manage their IT infrastructures. All this talk about simplifying IT really means something.

Thanks for the kind words this morning Bob!

Back in January, we collaborated with VMware and other storage vendor partners in a multivendor post to make sure iSCSI customers were successful in deploying iSCSI for VMware.  Dell has a close engineering partnership with VMware, as is EMC, to assure integration of storage platforms with vSphere via the vStorage APIs.

There have been some notable changes to VMware's architecture and how it works with iSCSI, so it was important for us to update you on the latest at the Dell TechCenter.

I LOVE coming together for the greater good!

I was scrolling through my blog reader and came across a post by Dave Rosenburg that piqued my interest: “KDDI chooses 3 Terra for cloud infrastructure.”  Having lived in Japan many moons ago I’m always interested in getting updates on what's happening in tech over there and since this involved the cloud, I was doubly piqued.

Gaijin Clouds gathering

Turns out that KDDI, the number 2 telecom provider in Japan (which makes them pretty humongous) has not only become a cloud provider as of late but have gone with gaijin technology to do so.  KDDI’s recently launched “KDDI Cloud Server Service” is powered by  3Tera’s Applogic cloud compute platform.  According to the 3Tera press release:

Initial offerings include virtual systems and virtual private data centers run at the KDDI Telehouse domestic data centers. This allows KDDI to offer both Infrastructure-as-a-Service and Platform-as a-Service solutions, where customers can run their existing applications on the IT platform or use KDDI’s prepared applications to significantly lower their initial investment and operational costs.

As Dave points out in his blog, its interesting not only to see a Japanese company  embrace the cloud but using outside technology to do so.

KDDI has made a big step forward, it will be interesting to see what the uptake is like.

Fun facts to know and tell: The word for cloud in Japan (kumo) is the same word for spider (kumo).  Now the characters used for both are different and the Japanese use the English word “cloud” when talking about cloud computing but still, I’m looking forward to getting the chance to present on cloud computing in Japan and make some bad pun involving the two.  Corny?  Yes, but that’s how I roll.

Pau for now….

 VM Squad posted today about a problem he perceives exists with Dell EqualLogic arrays. Its not a problem, but that's what one of competitors is telling people.  The issue is the time it takes to upgrade firmware in our systems, which is 15 seconds in the systems we are selling today. 

What's interesting is that Derek Schwab posted about the exact same thing yesterday and how it worked so well. 

So what's up with that?  

Systems and storage solve intermittent communication problems through SCSI time outs. If the host system can't communicate, it keeps trying patiently for a long time before giving up. This is a lot longer than 15 seconds.  The amount of time depends on the host system implementation, but it is usually more than a minute and can take five or more minutes (or so I'm told). FWIW, this is the same mechanism that is used for multi-pathing.  After a SCSI timeout, the system tries to re-connect using an alternative path.   

So a 15 second delay (not an outage) a few times a year is not a very big deal. If you have to, you can schedule it for non-peak hours. Everything will work, applications will stay up and end users will see a short temporary hang - if they see anything at all.