Enterprise cloud put to the test

Previous 1 2 Page 2

If you stay within the committed 5Mbit, you will only pay $125 a month. The extra charge comes in when you use more than your allocated bandwidth. Say you go over by 1Mbit for a total of 6Mbit, then you will have to pay 2x the Mbit fee (so $25 per Mbit would be $50 per Mbit for overage). Our total for the month would then be $175. Fortunately, Terremark allowed us to cap the bandwidth at 5Mbps for the VPN connection, which is all we used. According to the billing invoice our Committed Bandwidth was in the 5M to 50Mbps Tier but that does not apply to the VPN. The VPN bandwidth is a flat-rate per month based on connection speed and is not included in regular bandwidth calculations. They have the following tiers, 1Mbit = $200, 3Mbit = $550, 6Mbit = $1085, 10Mbit = $1285.

Overall, we liked Terremark's management app, and its speed to delivery. Provisioning was simple -- even though we did all of the virtual machines from the pool allocation allotted to us, and integration with our non-standard router was painless. We don't mind pain for gain, but it wasn't necessary with Terremark.

Rackspace

We were a little frustrated by Rackspace. Rackspace's process was slow, and may be faster for others as our negotiation and installation were done somewhat outside of their normal sales processes. The upside is that Rackspace's costs were more transparent and once rolling, their performance was very good. Rackspace provisioned us on Dell hardware, but emphasizes that most other top tier brands/models are available. We got the feeling that they're used to dealing on longer negotiation cycles with more diverse hardware needs, and deployment cycles associated with very large organizations.

Once the hardware and virtual machines had been provisioned, our site-to-site VPN took a while to integrate as well -- and much longer than the competition in our not-using-Cisco test. Once the VPN worked, it was smooth sailing, although IIS was installed on every Windows Server 2008 machine (we used Apache for testing), so we had to uninstall everything (IIS stuff) first. Some of these seeming disconnects could have been the result of our abnormal provisioning. The Red Hat Enterprise Linux virtual machines were correctly setup. As with BlueLock, Rackspace's virtual private cloud was fully provisioned on top of VMware ESX 3.5 by Rackspace so we didn't have to create the machines ourselves. There is a spot in the administrative Web interface to create new virtual machines (through a request), but this is limited to Windows Server 2003 and RedHat RHEL 3, 4 and 5. It's possible to have what you like (such as virtual machines), but you must submit a ticket for that with incumbent additional cost.

We liked the Rackspace administrative portal. The portal had six main divisions, Support, Products, Services, Network, Account (management) and Community. It's integrated, like Terremark's, and offers a tabular method of drilling down to support tickets, viewing each server resource utilization, viewing time series of performance characteristics, and administering our account. We found the Community tab interesting, as it took us to a private user forum. The forum is designed not be used for trouble tickets, rather communication among Rackspace clients for items like application integration, performance tweaks, and so on. This type of community-based communications was missing in BlueLock's and Terremark's offerings. It's like an internal user group.

Rackspace's communications with our n|Frame NOC was very fast, despite the long distance (Indianapolis to Austin) and we were happily surprised at the speed. Our ability to control VMs was also good, and we could manipulate our VMs readily although we couldn't actually connect to the console of the VM from an external (to the VPN) connection. It's also possible to review antivirus and URL monitors, but we didn't 'purchase' these. Interestingly, we could use the portal to buy SSL certificates (five types from VeriSign or two types from Thwate) -- very convenient, we thought.

We provisioned the Rackspace virtual machines for testing with our benchmark and connectivity tests. There were no mysteries, and Rackspace's Dell hardware performed well. We had no difficulties administering changes with Rackspace although gaps in their response were as mentioned, likely to have been the product of not being an actual customer.

We liked Rackspace and were it not for its slowness, we'd have liked the product much better, even though we know we were exceptions to their normal sales/fulfillment process. Rackspace's portal is useful, although with fewer choices than Terremark's and with a bit less functionality. As we seemed to have hurried them, we didn't get the full customer experience we were hoping for. Nonetheless, they were in the mid-range of pricing, and performed very well.

Costs

We asked each competitor to keep track of costs for us. Each competitor was a bit cagey and all three wanted to emphasize that costs are variable and tiered. They did, however, eventually get us pricing that reflected our utilization figures after we tested each private cloud with a performance analyzer to gauge CPU, bandwidth, VPN, storage, and other costs.

We also attempted to compare the three service providers with a do-it-yourself option -- in other words, buying hardware and software and deploying the apps on your own. With the comparison lies strong caveats. If one uses a DIY-type solution, there are hidden expenses involved that we didn't include in our estimate. These include support staffing, and leasehold costs, although we did include a collocation cost for power and space, at $45 per rack one unit per month prorated over the cost of the Dell hardware we chose in our DIY cost simulation. We also didn't include applications or application support, although these aren't covered by our competitors, either. Nor is the cost of negotiations, procurement, shipping, or building the hardware components included.

Our final caveat is that pricing appears to be a moving target, and a heavily guarded sales secret. The phrase "it depends" applies heavily here, as it became clear early in our review that we'd have to keep a taut line on our specification to allow even a close oranges-to-oranges comparison. And for those using virtual private clouds for availability, N+1 or 2N availability requires off-premises extensions of equipment, making DIY impractical.

Henderson is principal researcher and Allen is a researcher for ExtremeLabs in Indianapolis. They can be reached at thenderson@extremelabs.com.

NW Lab Alliance

Henderson is also a member of the Network World Lab Alliance, a cooperative of the premier reviewers in the network industry each bringing to bear years of practical experience on every review. For more Lab Alliance information, including what it takes to become a member, go to www.networkworld.com/alliance.

Read more about data center in Network World's Data Center section.

This story, "Enterprise cloud put to the test" was originally published by Network World.