Networking

Bandwidth allocation: Common-use or dedicated links?

When it comes to provisioning WAN bandwidth, if the legendary inch is given, a mile will be taken. IT pro Rick Vanover weighs pros and cons of dedicated links for certain business services.

Start talking with any cloud provider and if you go so far with any solution, bandwidth will surely become a topic. Take, for example, cloud-based storage where many customers who have gone ‘all-in’ on cloud storage have provisioned an entirely separate Internet connection just for the cloud storage route. The traffic to the cloud storage provider is routed explicitly through that designated connection. This also occurs at times with storage grids for data protection strategies where a dedicated link is provisioned for various storage systems that move data around an enterprise.

This leads to a pretty dividing question, should bandwidth be centralized and shared or split up into dedicated point-to-point links for the biggest consumers?

One side of the argument says to put all of the bandwidth in one connection in terms of the most aggregated throughput. And if the carriers can provide redundancy with the bandwidth, this can round out the resiliency of this approach. In this situation, the most total bandwidth can be enjoyed; but there is risk of one consumer taking too much of the bandwidth.

In situations where quality of service (QoS) is just not enough of a granular separation, separate lines for each primary consumer may make more sense. This can come about for many reasons, but costs and cost allocations may be one of the most common. Using an example, a simple way to put this is that if the storage grid consumes almost a connection of its own; then that group should fund their own connection. Take that one step further and ask yourself if you are paying for so much of the connection for your storage technologies, why would you want to have a QoS rule put into place?

I can go either way, and of course the ultimate answer is that it depends. Using separate connections for the largest consumers also allows for failover options to the other connections in some situations.

Where do you stand? Centralize the bandwidth and WAN connections, or split up connections for individual services that consume the most traffic? Share your comments below.

About

Rick Vanover is a software strategy specialist for Veeam Software, based in Columbus, Ohio. Rick has years of IT experience and focuses on virtualization, Windows-based server administration, and system hardware.

8 comments
Justin James
Justin James

Impossible question. What's my scenario? If I moved to a hosted Exchange solution, there's a good chance that I'd dedicate bandwidth to it, and if my company was big enough, I'd want frame relay directly to the vendor. If were were talking about a less widely used application, I'd probably say "no". Totally situational, and there's no way to ask a blanket yes or no question on it. J.Ja

jasta99
jasta99

A nice hybrid solution is running two circuits as Active/Active. Dynamically route general traffic over one and storage over another. You get redundancy for all traffic and not pay for unused bandwidth. The only time you have bandwidth contention is in a degraded state. Generally this is an accepted risk. The only real downside is the need for BGP multihoming. Outside of that, you grow into the solution you need. Typically you can start by sharing pipes with QOS and when the concept really takes off break off into separate circuits as needed.

gregaltman
gregaltman

We took the route of dual connections, one for internet use and one for interbranch WAN. This keeps the web surfers off our WAN. and provides a backup connection in case our WAN link has a problem. We have 7 offices totalling about 150 people.

mogulsurf
mogulsurf

The first thing that comes to mind when I read your article is "What's the size of the organization?" We're a small-medium size company with 130 employees and I can't imagine provisioning separate links for various services. Could you comment on the representative size organization (if any) in your article?

tracy.walters
tracy.walters

My company has four offices and about eighty employees. We have a 10MBPS primary and 3 MBPS backup into our two primary offices. Those offices have our server/SAN suites in them, one is primary and one set up for Disaster Recovery Failover. Our smaller offices have Bonded T1 lines for access. Through MPLS and QoS we use our lines very efficiently, and our carrier is very proactive regarding monitoring of our links. We use Citrix for most of our work, which means we don't have huge downloads going on across our network, unless someone needs a file downloaded for field work. Setting up your network for efficiency takes some time in planning and some monitoring while in place, but it pays off huge dividends in personnel not even being aware they are remote from the servers they are accessing, and when they need an application, it just works. With our virtual servers/Citrix servers, we build those boxes to be very robust...putting lots of processor and memory in them means faster response times for queries, and more efficient use of their time. A few thousand dollars in the infrastructure means those expensive people, especially if they are billable like ours, can make you a lot of money fast.

b4real
b4real

There are a number of solutions that can provide multiple connection balancing for the SMB. The larger enterprise its a little easier - but this doesn't entirely revolve around organization size.

BroadcastArashi
BroadcastArashi

I've read the article, I've read the replies, and I still don't see the logic in dedicating links for storage or any other application when you can simply use QoS effectively. Bandwidth is cheap, even more so in the United States. I think it is most logical to keep it simple. Have 2 equal bandwidth links. One for primary use, one for backup in case of primary failure. Usually you can get some kind of deal with your ISP on a backup link for redundancy purposes. Use identical QoS on both links to allocate bandwidth in the case of congestion. If other protocols are not contending for use of the link, storage can use as much as it pleases. The advantages of this are simple design, easy troubleshooting, and resilient failover. For better redundancy, you can even configure this scenario with one router controlling the primary connection, and one router controlling the backup connection. However, the only problem with this is if the redundant link is not priced as a backup link, then you may want to load balance to achieve higher throughput by say putting storage on one link, and everything else on another, and having each link failover to the other. It's more complicated, because you need to manipulate BGP to prefer one link for storage inbound and outbound. If your ISP supports it, you can use multilink PPP to join the links together to utilise all the bandwidth. Say two 5mbps links could become a 10mbps link. This would mean that you couldn't have redundant routers as far as I can tell, but maybe someone knows how to do this. This might be a better solution. But I have no idea how you'd configure QoS to work with the link at both 5mbps and 10mbps without having to manually change the QoS policy. Any ideas on that? You could just convert the absolute values into relative values, but for some protocols you may need a fixed value, such as for VoIP.

brettman
brettman

I agree in principle with all that has been said. A single pipe with Qos (not router based qos something with tcp rate shaping and advanced UDP control) is all that is needed in most cases. A link of any size can be carved up into granular junks for precise control of all applications with bursting so there is never any bandwidth wastage. ie: one 20mb link with traffic shaping is better (in terms of control) than two 10Mb links without. However like everyone has been saying , there are benefits of redundancy in having two links. A bandwidth management device worth its salt can fit seamlessly into either an active/standby or active active configuration. Then there is also the issue of cost. In many parts of the world bandwidth is the single largest recurring IT spend. I'm of the opinion that in most cases whether you have single or dual links bandwidth management / qos is helpful in getting the maximum potential out of the given bandwidth.

Editor's Picks