Desperately Seeking Virtual I/O - Eye on Blades Blog: Trends in Infrastructure -
Eye on Blades Blog: Trends in Infrastructure » Desperately Seeking Virtual I/O
Desperately Seeking Virtual I/O
Our friends at Dell are at it again.  In case you don’t pay as much attention to the happenings in the blade world as we do, here’s a recap.  Dell has been chasing the HP BladeSystem for some time now.  Though bamboozled on several fronts, the searchmetal_detector_convention for a competitive virtual I/O solution to Virtual Connect has been the most elusive.  Here’s the retrospective from March 2008 to February 2009. 

1. Virtual I/O ‘Take One’ for Dell was a partnership with eGenera in March 2008.
Headline: eGenera Inks OEM Deal with Dell

“Dell is listening to customers and providing solutions that make the virtual data center easier to deploy and manage, regardless of platform,” said Rick Becker, vice president, Dell Software & Solutions. “Dell and Egenera will help customers focus on company growth by delivering excellence in virtualized infrastructure from server performance, storage interoperability to dynamic data center management.”

 

 2. Dell asked for a mulligan in July 2008 when they tossed eGenera overboard and tried to create something sort-of like Virtual Connect; meet FlexAddress.  Over a year and half late to market after version 1.0 of Virtual Connect.
Headline: Dell joins ranks of I/O virtualization providers with FlexAddress

"We have taken a very different approach than HP. Theirs is a proprietary switch that plugs into their backplane, and after adding all of the switching, it can cost about $20,000 for one chassis," said Rick Becker, vice president of software and solutions, Dell Product Group. "We have done this using open standards, so FlexAddress works with other switches like Cisco and Brocade and users don't have to switch their switches."

3. Dell and go back to the partner route with a Cisco bear hug. Headline: Dell and Cisco team up on next-gen datacentres

Sorry, no quote from Rick Becker on this one, but here’s another that will give you an idea.  "We're hearing much more interest from customers on FCoE, even though it is not an officially ratified standard, but at the moment, iSCSI is available and affordable," said Robin Kuepers, head of storage for EMEA at Dell.  No one asked if the Cisco's Nexus switches that Dell has simply agreed to resell are compatible with Cisco's large installed base of Catalyst switches or all the other industry compatible switches such as ProCurve and others.  Hint: They're not. (UPDATE: The comments below called me out that this was overkill and they are right; I should explain. There are differences in feature/functions between Nexus and Catalyst, especially related to capabilities with VM's and the future implementation of FCoE.  Catalyst switches will be missing out on some of these capabilities under the current direction.  The right direction is to have open standards for these types of capabilities, regardless of the plumbing in the datacenter and a Nexus-only approach doesn't get us there either.)

4. That brings us to Tuesday, February 3rd and a 180 degree sprint, to yet another partner.
Headline: Dell Pairs with Xsigo on Virtual I/O 

"What we are excited about with the Xsigo appliance is the openness," explains Rick Becker, vice president of software and solutions at Dell's Product Group. "This works across form factors and vendors, and in a heterogeneous data center, this can mange it all no matter what logo is on the box."  No shortage of handshaking going on over there. After all, HP's Solution Builder Program has over 300 members.  

However, there’s one important thing that you should know about the Xsigo Dell announcement that Dell didn’t mention.  It does its magic with InfiniBand. That means you need to have an Infiniband network in place to use this product.  I’m all for open standards in this space too, like Anthony Dina at Dell blogged about last week, other than companies that have decided to base their server interconnect on InfiniBand, there is a more straightforward path for customers looking for a datacenter-wide virtual I/O strategy and an end-to-end virtual infrastructure.   

We do agree with Dell that not all businesses are the same, so a ‘one-size-fits-all’ solution isn’t a good approach.  To complete our offerings, HP partners with Xsigo among other virtual I/O vendors.  It’s supported on HP blades and rack/tower servers too.  However, our cornerstone is advancing innovation like Virtual Connect Flex-10 on plain old fashioned Ethernet and Fibre Channel that most businesses will find is a ‘flex-size-fits-most’ solution that fits them just right.
Posted 02-10-2009 3:11 PM by newtonja
Filed under: , , , , , ,

Comments

James Nelson wrote re: Desperately Seeking Virtual I/O
on 02-10-2009 4:50 PM

I think it might be useful to understand the Xsigo solution better.

The Infiniband piece is just connectivity fabric which is reliable. The operators on the server side see something that is familiar, easy to configure and simple -- virtual ethernet and fibre-channel interfaces.

Add many of these Xsigo virtual interfaces based on demand, bind them uniquely to your VMs, set QoS to guarantee bandwidth and performance, move these IOs from VM to VM, or server to server without ever re-wiring your rack.

This is quite sweet and one gets server utilization to pretty high numbers - as you have freed up the IO bottleneck.

Think of this as a converged fabric - without ever needing to throw away your ethernet and fibre-channel switches (unlike the new fangled FCoE). Xsigo is great for investment protection and consolidating your infrastructure - perfect for these difficult budget situations and tough economic conditions.

Jason Newton wrote re: Desperately Seeking Virtual I/O
on 02-11-2009 12:07 AM

I'm there with you James.  

Just in case my post is misunderstood, I really wanted to avoid slinging arrows at Xsigo.  I you are already doing Infiniband, Xsigo rocks and if not, it still makes sense for some.  From a big picture view, you still have to manage the Infiniband fabric plus the Ethernet and Fibre Channel fabrics.  

Obviously I was really poking on Dell's I/O strategy - not on Xsigo.

We definitely agree that FCoE in some implementations needs to include a giant forklift in the price for the upgrade.

Jon Toor wrote re: Desperately Seeking Virtual I/O
on 02-11-2009 4:54 PM

Xsigo offers another unique benefit beyond what James mentioned above: it's 100% open.

Xsigo virtual I/O works with both blade and rack systems from every X86 server vendor, across most operating systems and hypervisors.

This is is huge win for management simplicity.

Data centers are almost all heterogeneous. Single vendor virtual I/O solutions therefore do not address the real need to consolidate  management.

Because Xsigo works across all platforms, IT managers gain a single view of the I/O infrastructure. Which is what's needed to cut management costs and increase resource utilization across ALL devices... not just on the subset that support a particular solution.

Regarding Infiniband, IB is today the ideal fabric for virtual I/O. Why? Aside from being fast,  power-efficient, inexpensive, and enterprise-proven, IB has one other key attribute: it's available everywhere. Every major vendor offers IB connectivity for their blade and rack systems.

This wide availability again fufills the mission of Xsigo being an open solution. Manage everything from one place, with no rip and replace. This translates to simplicity and cost savings, benfits that very much align with today's IT realities.

James Henry wrote re: Desperately Seeking Virtual I/O
on 02-18-2009 3:23 PM

All valid points guys but ultimately it all comes down to costs and that is why Virtual Connect has been so successful as it costs roughly the same as equivalent LAN and SAN switches and it just plugs straight into customers existing backbones while providing all or at least most of the benefits of virtualised IO. Infiniband has been bouncing around for such a long time now - 5 years ago we were all talking about it being the next server I/O standard but PCI eXpress came along and Ethernet continues to prevail as the preferred common fabric. 10Gb NICs are already appearing in servers and innovations such as Flex 10 NICs and VC Flex 10 are helping customers leverage the advantages of 10Gb while driving down costs. Most customers still find the costs associated with Infiniband just too high too stomach.

Jerry wrote re: Desperately Seeking Virtual I/O
on 03-01-2009 8:32 PM

Just curious about #3 above in which you state that the Cisco Nexus and Catalyst switches are not compatible.  I'm running them together and they certainly interoperate just fine.  Would you care to expand on what makes the Nexus incompatible with Catalyst and ProCurve?

Brad Hedlund wrote re: Desperately Seeking Virtual I/O
on 03-01-2009 11:58 PM

The statement in this article that Cisco Nexus switches not being compatible with other Cisco Catalyst switches, or any other standards based ethernet switch is patently FALSE.  

Thanks for the laugh :-D

David Fitzgerald wrote re: Desperately Seeking Virtual I/O
on 03-02-2009 9:22 PM

Virtual Connect offers very little to virtualization issues, I don't understadn why you are making ushc a big story out of it.

Infiniband has been there for ages, and today it is only used in some HPC clusters, even if Xsigo is willing to sell their invention at a reasonable price, we need to compare the prices of IB and FCoE adapters.. FCoE is definetly the future./

newtonja wrote re: Desperately Seeking Virtual I/O
on 03-04-2009 3:49 PM

Mia Culpa.  Saying that Nexus and Catalyst are incompatible"is overkill  on my part.   I should have taken the time to explain the issues I see.

Technical fact - Nexus switches and Catalyst switches work fine with each other with 2 major feature exceptions.

Both can run in today's industry standard Ethernet, but tomorrow's Ethernet will include 4 new specifications that will upgrade the Ethernet network to be more efficient for carrying block storage traffic like Fibre Channel over Ethernet (FCoE) packets.

Nexus is designed to implement the new Ethernet (generally called "lossless Ethernet"; industry terms are CEE or DCE) so that it can carry storage traffic using FCoE packets. Catalyst is not designed to support this.  The IEEE and CITS standards needed to implement CEE are not ratified yet, but they should be.

When connected to VMware, Nexus offers some special network management advantages for VMs. This is currently dependent on a special technology called VN Tags. Today, it is Nexus-proprietary. Other switches including Catalyst, ProCurve, Virtual Connect (not-a-switch) can work in a Nexus network but some of the special VMware functions won't work.

Robert wrote re: Desperately Seeking Virtual I/O
on 03-14-2009 2:43 AM

So will HP update Virtual Connect to seamless pass the VN Tags from the software Nexus 1000v to a physical Nexus 5xxx switch? If not, then it's a rip and replace of the VC modules which is major money down the drain.

Add a Comment

(required)  
(optional)
(required)  
Remember Me?

Type the numbers and letters above:
Powered by Community Server (Non-Commercial Edition), by Telligent Systems