In-Depth
The Great Cloud Bottleneck: How Capacity Issues Can Kill Your Cloud Project
Cloud computing vendors promise the world, but if your network isn't up to the task, the resulting speed, bandwidth and latency issues may doom your project from the start.
More on this topic from Redmondmag.com:
Ever fall half asleep waiting for a Web site to refresh or a Web app to respond, or freak out when your cloud mail is down? If so, you understand a fundamental problem with the cloud: Its speed utterly depends on our network connections, be it a WAN access point, the Internet or a private line.
So we're not all confused with varying definitions, let's agree on what a WAN is. According to Cisco Subnet, an independent community for Cisco users, "The WAN is a place in the network that aggregates various types, speeds and links running a disparate set of protocols together crossing metropolitan, state and even country boundaries. The largest example of a WAN is the Internet itself, which can be regarded as the public WAN. The primary purpose of a WAN is to connect users and applications connected to various LANs."
WANs and the Internet were getting clogged long before the cloud. Multimedia, Web and video conferencing, surfing, VoIP and unified communications already stress those networks that haven't been thoroughly boosted.
IT pros find these clogs frustrating. "I find it truly amazing that the T1 speeds that were to dream for years ago are still the standard for many links," says Scott, an IT officer for a savings bank. "While all other technology has boarded the enterprise for journeys to other worlds, the standard WAN speed has stayed grounded. It's not just the SaaS [Software as a Service] applications, cloud solutions, Web sites, downloads, media streams that call for higher bandwidth. It's centralized backups, data transfers, SAN replication and so on, that cry for more bandwidth. Unfortunately, the price tag rises significantly. Oh, and if you have VoIP [Voice over IP], you need to guarantee a portion of that bandwidth for the voice traffic." Scott notes: "there are some lucky souls out there that have the money to put fiber all around, or even large pipes through Metro Ethernet, but many of us still are bound by budgeting dollars."
Public cloud services have some of the biggest bandwidth pains. A high-level executive recently confided to me just how maddening it is to use Salesforce.com. The experience soured the veteran CEO and entrepreneur on the cloud. It even slowed down his company's cloud product development after he heard similar concerns from customers.
Some have already seen slow real-world performance. "I've just put my first customer on the cloud using Office 365. Their first comment was, 'Why is it so slow?' But for $6 per user per month [P1 plan], that's a bargain compared to Business Productivity Online Suite [BPOS]," says reader Ian Guyer. "I'm not sure the cloud will ever be as big as anyone hopes. Not a single customer wants to go slower. Ever. Go ask any computer user if slower is OK and see what they say. Opening a large PDF on the cloud can be agony."
Good or bad, the cloud forces IT to come to grips with all aspects of the network, internal and external. "I'm beginning to think that the whole 'cloud' terminology has been thrown around so much it doesn't mean much anymore," says Dustin Harper, head editor of MSTechPages.com. "The cloud is the WAN. It's common sense for anyone that has worked in IT for more than a few years -- before the 'cloud' was a buzzword -- if you have a slow WAN link, anything utilizing that link will be slow. You can connect to the world's largest and most powerful supercomputer running an OC48 with a 56k dial-up link, but you'll never get any benefit of the speed from the provider unless you upgrade your weakest link: your WAN connection."
Vendors on Cloud Barriers
Virtualization vendors, which form the infrastructure of cloud services, are concerned about bottlenecks and working like mad to clear them. Speaking to Redmond sister publication Virtualization Review Editor in Chief Bruce Hoard, Simon Rust, vice president of technology at AppSense, had this to say: "Infrastructure availability is concerned with the availability of suitable, resilient, always-on network infrastructure between the end user's endpoint device and the cloud computing resource. Some cloud-delivered applications can cope with highly latent connections between the two points, since they make use of HTTP connections, which were purposely designed for long-distance connectivity with potentially poor links," Rust says. "However, as we find ourselves looking to deliver services that have a much lower tolerance to latency -- desktops, for example -- then the network link between the two points becomes critical."
Rust continues: "Infrastructure availability is an area that is a little more complex, in that this really comes down to network capability at the endpoint. As the consumerization of IT continues to make things smaller, cheaper and more attractive to consumers, the network capabilities of these devices become more and more necessary and, ultimately, critical. These devices require always-on connectivity, and on today's connectivity, only the applications capable of handling high-latency connections are appropriate to these devices.
"While there are available options to the mobile business user while on trains and planes, these are similarly unable to operate with applications that require anything other than perfect, latency-free connections. We must wait for the infrastructure to be always on and to exhibit extremely low latency."
F5 Networks Inc. CTO Karl Triebes also spoke with Hoard, asking a number of probing questions: "One of the most challenging aspects of the cloud deployment model is that it is dependent on a WAN. Bandwidth and latency issues are inherent with a WAN and can cause application migrations to fail repeatedly. For those applications that are successfully migrated to the cloud, how are they made known to and accessible by the infrastructure so users can be directed to them? How will cloud-based applications ultimately perform across a WAN?
"For many organizations, these are still unknowns."
WAN-performance fears, or a reluctance to spend a fortune on a speedy network, have many in IT holding back, to put it lightly. "With a local datacenter, all resources are, well, local. With the cloud, you require a complex infrastructure -- local, the ISP, the Internet and the cloud provider -- in place to even get to your data," explains reader Craig.
"I get to rent my software, and pay higher monthly ISP bills, and get to be at the mercy of outsourced IT issues? Where do I sign up ...?" asks one reader, who did not want to be identified.
Reader Mike from Southern California is likewise skeptical. "I'm still dumbfounded how the hype machine over cloud computing continues to roll forward. I believe that the cloud should be another tool in your toolkit," he says. "Like any other technology, you need to understand it, but you also need to know how the cloud fits into achieving your business goals. Don't create a business case just so you can use the latest fad."
Real-World Scenarios
One reader believes that IT is often caught by surprise by WAN issues. "No one seems to anticipate the need or cost of additional Internet bandwidth and redundancy," explains reader Dan S., an IT specialist for the federal government. "For example, let's take a mission-critical application such as e-mail and move it to the cloud for a company of 5,000 people. In this scenario, prior to moving to the cloud, the e-mail system would continue to work internally and users could get access to e-mail data even if the Internet connection became congested or went down. Once you move that application to the other side of the Internet pipe, the importance of the company's Internet connection goes to a whole new level.
"Generally, the existing Internet pipe doesn't have the sufficient bandwidth to support the new load of using the previously internal mission-critical app in the cloud. Some companies have gotten around this by deploying QoS [Quality of Service] and prioritizing certain types of traffic, but this doesn't always work well in a cloud solution. Quite often, companies that move to the cloud find themselves needing to beef up their bandwidth just to support the use of external applications that were previously in house. This situation is only compounded by the latency issue with existing internal WAN congestion," says Dan S.
Midsize companies have it worse, Dan S. believes. "Typically, midsize companies have a single pipe -- whether it's a single connection or an aggregation of connections -- and they generally deal with Internet outages as an annoyance versus a work stoppage. Once the mission-critical applications move to the cloud, the Internet connection becomes a single point of failure and, as such, requires redundancy to be added," Sheehan continues. "I make it a point to discuss the importance and 'hidden costs' of the Internet connection to support cloud applications with everyone who mentions they're considering the cloud."
WANs Done Right
Cloud consultant Mike Goodenough has plenty of questions and more than a few answers for those in IT considering clouds. "How is each site going to get access to the cloud environment? How are the on-premises systems going to communicate to cloud applications? How much bandwidth is going to be needed between each site and the cloud provider?" he asks.
Figuring out bandwidth needs takes a lot of elbow grease. Here's how Goodenough approaches his clients: "We need a diagram of how every site communicates today -- is it across the Internet or dedicated MPLS circuits? We need to find out how many different internal networks are used and how each application communicates between them. We need to measure WAN utilization between sites and the main datacenter where the applications reside. We also need the size of each site's Internet connections," Goodenough explains.
"We would likely need to assess the availability of the WAN infrastructure -- look at redundancy, availability, SLAs and bandwidth of the WAN, and potentially assess the impact on business services in the event of an outage."
Bigger shops have more complexity to deal with. "The larger the enterprise, the more complicated the network is going to be. The larger enterprise environments will have more dedicated MPLS circuits instead of relying on the Internet connections," Goodenough says.
Some WANs are ready for the cloud, and some aren't. "Sites that are already using Internet connections for all their WAN connections are ready to move to cloud. The WANs that are using dedicated Frame Relay or MPLS circuits might need to look at upgrading their Internet circuits to start using public clouds," Goodenough says.
Not all cloud apps have to be bandwidth hogs. "If you move a complete application stack to the cloud, bandwidth should not be an issue. The main consumer of bandwidth is the communication between application servers, not end users back to the application. All that server-to-server communication should occur on the cloud provider's LAN and only the end-user access will go over the WAN," Goodenough explains.
The Cloud Doesn't Cover Everyone
Phil Mahon, of Canadian-based Cottage Computing Inc., couldn't move to the cloud if he wanted to -- he's too remote. "Technology has permitted many of us to move away from urban centers and live in more rural, pleasant situations, without commuting. Problem is, try and solve the last-mile problem. You can, by fluke, be near a location where DSL is possible, or even cable, meaning you can live with 5Mbs or less. But for most of us, we suffer with nothing except dial-up, or flaky wireless point-to-point radio maxing out at 1Mbs when it works."
Assuming remote areas get high speed, Mahon figures that will get maxed out in no time. "I fear the results if everyone assumes all is well and that WAN speeds are sufficient to, say, access a database. It isn't going to work any better for us than it did when all the urban centers got high speed. All the Web sites started putting junk images on their pages. It became intolerable, as will the cloud."
It turns out that too many of us don't have access to fast-enough lines for cloud apps. Reader Fred Linton is a case in point. "My 768KBps is a truly third-world-nation WAN speed, but that's what AT&T/Yahoo DSL offers in my neck of the woods -- when it's not dragged down to something closer to 450 KBps. For half my monthly DSL bill, I can get 3.6 or 7.2Mbps HSDPA [High-Speed Downlink Packet Access] service, but that's only when I'm visiting Poland -- not a chance of that happening here in these benighted ... err ... United States," Linton complains.
Reader Travis Stock feels similarly shut out. "This has been my concern from the beginning. I'm a one-man shop for an international construction company, but we are based out of Montrose, Colo. Not the smallest rural town you could come across, but on the western slope of Colorado, we're somewhat remote. There is fiber out here, but none that has been packaged and sold -- that I know of. If we ever move some stuff to the cloud, I had better have a very good reason, as on day one I will hear, 'It's pretty nice, but a little slow.' And as more users get on, the slower it will be," says Stock, an IT manager for San Juan Construction.
"You couple that with the rising demand for cheap voice [VoIP] and video conferencing -- both of which we've deployed -- and you're looking at the potential strangulation of a business-critical app that you moved to the cloud to make more robust."
Stock isn't that confident in network gear solutions. "What's somewhat humorous about the cloud is, you'll be working over the WAN, which is what WAN optimizers were made for, and yet, they will be very difficult -- if not impossible -- to deploy. We have Riverbed Steelheads deployed in our organization. So the one situation you could really use it, it won't be available."
Cloud Capacity Conclusions
In 1996, Bob Metcalfe, inventor of the Ethernet, predicted a "gigalapse" in a column in Infoworld magazine. A gigalapse is where the Internet becomes overloaded and falls apart -- leading to massive outages. Metcalf promised to eat his column if the Internet didn't fall apart, and eat it he did literally. While the Internet has been gratifyingly robust, there are questions as to whether it can keep up with ever-increasing demand. Reader Roger O'Daniel doesn't think the 'Net is up to the task, that the supply of addresses is an issue and that ISPs aren't investing enough in bandwidth (see "Cloud vs. WAN Costs").
Cloud consultant Goodenough begs to differ. "Most service providers today don't operate at capacity, and I would suspect it will be more of a steady growth rather than an explosion. The reality is that cloud adoption and migration is a fairly complex process that's more gradual, rather than just flipping a switch. I think service providers are anticipating this growth, and are growing appropriately," Goodenough says.
More on Cloud Computing from Redmondmag.com: