"You can't write an SLA [service-level agreement] for the Internet," said Tom Nolle, president of CIMI Corp, an independent consultancy firm for service provider and enterprise networking products and services. "It's a best-effort service, and that's all it's ever going to be."
On a private enterprise WAN that delivers apps to branch offices via an MPLS service, it's like a "direct flight," Nolle said. Packets take a predictable path and have fairly static latency and packet loss.
But assessing SaaS performance by those benchmarks over the Internet is nearly impossible since the parameters constantly fluctuate, he added. Packet transport becomes more like a flight with multiple connections -- slowed down every time the packets have to change carriers or hop to get to the vendor's data center or back to the user.
Although service providers have agreements to provide peering points with each other's Internet backbones, these relationships vary, Nolle said. Operators also often have little financial incentive to optimize their competitors' traffic flow at those connections.
"You're not going to find any operator who's going to sign a contract and guarantee you're going to get this number of hops," he said. "The only thing you can do is play the odds a little bit. Your best bet is to watch for signs of access congestion and fatten your pipe a little bit."
Cloud providers will probably have multiple data centers to alleviate WAN/cloud latency problems, but SaaS performance is still a gamble, according to John E. Burke, principal analyst at Nemertes Research.
"You have no control over performance characteristics if you access them over the Internet," Burke said. "There's not a lot you can do to improve performance [of your link] to a SaaS vendor unless they specifically allow some kind of private network connection to them."
Bandwidth unlikely to improve WAN, cloud and SaaS performance
Because the manner in which packets traverse the Internet is inherently fluid, leasing more T1 lines is going to have limited benefits, according to both experts. Unless enterprises are maxing out their existing links, a faster on-ramp to the Internet won't affect how packets travel between SaaS vendor and enterprise data centers, Nolle said.
"Buying more bandwidth is only going to help you if access bandwidth is your problem. You can't buy Internet bandwidth," he said. "The only thing you can do is buy the on-ramp."
There are still a few steps that enterprises can take to get at least a rough idea of SaaS performance at various locations across the WAN before committing to the cloud, according to Nolle and Burke.
Basic WAN/cloud performance testing, such as pings at various times and locations, may help determine average latency and packet loss, Burke said. Running traceroute commands a few times may also help determine which paths packets might take but offers no promises, Nolle said.
"Knowing that all of [those results are] impermanent and that, on the Internet, [performance] gets wildly different on a different day … you've at least got a shot at analyzing performance expectations," Burke said. "But it's in no way a guarantee."
If enterprises' Internet links are performing well today, both experts agreed, most SaaS apps on the market now shouldn't pose a significant problem. Basic client-server transactions common to SaaS aren't going to look much different from casual Internet traffic, but cloud-based storage services will be the least tolerant of latency and jitter, they said.
Public cloud-based SaaS performance may be akin to how remote users experience on-premise applications via an Internet virtual private network (VPN) client, Nolle said.
"These [cloud-based] applications are not high-volume enough to have any overall impact on traffic," he said. "They don't create a QoS [quality of service] problem, but they make a QoS problem that had already existed -- and existed all along -- a problem and not just an annoyance."
Can WAN optimization solve SaaS performance over WAN and cloud?
Caching and compression features in WAN optimization controllers might sound like a promising solution to improve SaaS performance, but there's one big problem -- an enterprise can't force a cloud provider to use the box at its end as well.
Even if enterprises could deploy symmetric WAN optimization with cloud providers, compression "nearly always introduces delay … and it may or may not help with latency and packet loss," Nolle added. "The value of [caching] is also incredibly difficult to determine."
WAN optimization controllers that inspect packets to prioritize certain kinds of traffic may also fall short because SaaS traffic is usually encrypted, according to Mark Urban, senior director of product marketing at Blue Coat Systems, which promotes the ability to decrypt, accelerate and re-encrypt packets with its PackerShaper product line.
All of these SaaS performance problems are compounded when enterprises backhaul Internet traffic over the WAN from branch offices to headquarters, Urban said. The additional trip makes SaaS performance more vulnerable to lags.
More than half (54%) of IT pros said they use two or more SaaS applications or other cloud-based services, according to an informal survey from Blue Coat, which polled about 150 IT and networking professionals at Interop 2010 in Las Vegas. Nearly a third (32%) use five or more SaaS apps or cloud services. Meanwhile, 68% of enterprises polled said that they backhaul Internet traffic from their branches through their data centers.
Targeting this WAN/cloud delivery dilemma, Cisco Systems claims its Wide Area Application Services (WAAS) controller can overcome backhaul latencies that affect SaaS performance by recognizing popular cloud-based services, reducing duplicate data streams and delivering 5:1 compression.
Let us know what you think about the story; email: Jessica Scarpati, News Writer