Data center fabrics have been in the news, but what does this term really mean?
We asked this question at a recent VMware users group. You can see the responses below. There was not much consensus.
This isn’t surprising. There is no single industry definition around fabrics.
But there are some consistent themes. I’ll highlight three areas where the various vendors pretty much agree. I’ll then point out three additional features where Xsigo adds unique capabilities that provide a distinct competitive edge.
Problems a Data Center Fabric solves
Before jumping into that, what are the basic problems a fabric is intended to solve?
Fabrics address the networking challenges introduced largely by virtualization. Traditional networking simply was not designed for the new requirements.
Areas where today’s networking approach falls short include the following:
Flexibility: When we start thinking of compute resources as a cloud, we need a lot more flexibility than today’s networks deliver. You want to decouple the service from the server, something that’s hard to do in a physically complex data center.
Cost: Multi-layered network hierarchies are expensive. When you compound that with the cost of additional storage networks plus large amounts of cabling to each server, you have an expensive mess.
Performance: Running data up and down network tiers adds latency and creates bottlenecks introduced by the Spanning Tree protocol. There used to be a simple answer for this: the three-tier data center. In that architecture there was just one hop from one server to another. But it was also a rigid design. It does not meet the needs of virtualization.
In these three areas, you can quickly sum up the benefits a fabric offers: it’s simple, fast, and flexible. With a fabric, you can achieve up to 10X improvement in performance, agility and reduced complexity. (While most vendors would agree with that, Xsigo would add that you can achieve those benefits with an incremental approach that involves little or no disruption to your current infrastructure.)
How a Data Center Fabric is Different
A fabric is a flexible network infrastructure. It typically offers three fundamental improvements over traditional architectures:
Flatter architecture: The first concept is to reduce network tiers. You have top-of-rack switches adjacent to your servers. We want to route data moving from one server to another entirely at that layer, rather than up and down a stack.
Fewer bottlenecks: One assumption is that a network faster than 1G will be employed. But beyond that, we also want to eliminate the bottlenecks that accompany the spanning tree protocol. Because this is a shared infrastructure, we also want to assign bandwidth controls to make sure the critical apps get the speed they need.
Flexible: Routing data around the infrastructure is tedious when you’re assigning VLANs and configuring ports at every step. With fabrics, you select a source and a destination and let the fabric do the routing. You do not have to worry about data isolation or high-availability. Those capabilities are built-in.
Those are the generic views of a fabric, and are consistent across most vendors.
What Xsigo’s Data Center Fabric Adds
At Xsigo we’d add a few more fabric characteristics that we believe are just as critical to meeting your ultimate objectives:
100% convergence: We believe a true fabric needs to carry all networks and all traffic types (and at their full line-rate speed). This sounds like something that everyone should accommodate, but that’s often not the case. Many times there are limitations:
- If you need to run more than two physical I/O connections into each server, you have not converged
- If you cannot run 10G Ethernet and 8Gb FC simultaneously at full speed over one wire, you have not converged
- If you have not consolidated, at the server level, all physically isolated networks to one cable, you have not converged
No scalability limits: If the term “VLAN exhaustion” is part of your vocabulary, a true fabric should allow you to eliminate it. If you are limited on the number of servers per fabric, you should not be. And if you cannot run as many VMs per server as you’d like due to bandwidth limits, a true fabric should remove that limit.
Fully open architecture: You should have the ability to pick software and components based on your needs. From servers, storage and networking to hypervisors and operating systems, to bare metal vs virtualized, on a true fabric your options should remain open.
Only Xsigo meets all the requirements
Only Xsigo converges Fibre Channel and Ethernet without restriction. On performance, interoperability, and reliability, the Xsigo FC performs as well or better than conventional FC. With Xsigo’s 40Gbps highly reliable fabric, dropped frames don’t happen. Throughput is never compromised, even when 8Gb FC and 10Gb Ethernet share one pipe. And interoperability is guaranteed by industry-standard silicon.
Only Xsigo removes scalability limits. There is no reliance on VLANs to create isolated connections. And the fabric can grow aggregate bandwidth and ports to accommodate 1,000 servers.
And only Xsigo offers an open architecture designed to work with all X86 servers, networks, and storage. Xsigo presents standard NICs and HBAs inside the server, and presents standard silicon to the LAN and SAN, thus ensuring interoperability.
So, we’ve listed six fabric attributes. Three of them you can achieve with multiple vendors. (Though we’d again maintain that Xsigo gives you the least disruptive way to get there.)
All six capabilities are available from Xsigo. Your needs will dictate what works for you.
But Xsigo should be part of your comparison, regardless. While the competition bats .500, Xsigo hits the home run. But that’s to be expected. Our solution competes and wins against the biggest players in the game. It has to be better.
