It’s said a picture is worth a thousand words, so I thought it prudent to begin my post with a photo of the network rack powering Voxel’s latest VoxCAST deployment, our Hong Kong Point of Presence, or “HKG1” as we know it.  Certain of our Asian VoxCAST end users will begin to see their content served out of this location in the coming days…

Our deployment consists of  VoxCAST servers and core network equipment inside the Equinix Hong Kong (HK1) datacenter, with infrastructure at the iAdvantage MEGA facility at 399 Chai Wan Road (commonly known as “Mega-I”), to facilitate additional peering and transit connectivity.  This POP connects to our Singapore (SIN1) datacenter, as well as to our US and Europe presences.  We’re also peering at the Equinix Exchange in Hong Kong, with expansion plans including a port at the larger Hong Kong Internet Exchange (HK-IX).

Part of Voxel’s goal when we announced our Asian expansion plans in late 2008, was to bring rich connectivity and interconnection to our hosting and delivery customers in Asia, just as we’ve been doing for 10 years in our US and European locations.  With a much different (and challenging!) network landscape in Asia, we’ve worked diligently to tackle routing and delivery issues that exist due to the complicated nature of nationalized telecoms in the region.  Expanding our network to Hong Kong represents a major step in our capabilities in Asia.

Why Hong Kong?

A mere thirty (30) milliseconds away from Singapore, Hong Kong provides access to a wealth of additional service providers, along with an evolved carrier ecosystem.  PeeringDB (https://www.peeringdb.com/) provides the casual onlooker with a good listing of providers and peerings in each market.

Benefits to our Voxel Asia Hosting Customers

As our Hong Kong and Singapore locations are directly connected, VoxSERVER and VoxCLOUD customers in Singapore will be able to take advantage of  our rapidly-expanding peering footprint in Hong Kong, with several major access providers in the Asia region pending activation (we’d love to tell you who, but pesky NDAs are getting in the way).  Our Hong Kong expansion provides our Singapore server customers with the best of both worlds — they are now provided with unprecedented regional (and global) reach, while at the same time maintaining a strong local presence, including access to our Singapore office’s world-class support and account teams.

Noteworthy will be the value proposition of hosting and delivery content to Chinese Internet users — our “gateway into China” in Hong Kong will provide better access to the  mainland market, while keeping content and intellectual property hosted in the strong, business-friendly country of Singapore, where copyright and intellectual property laws are well understood and enforced.

A Commitment to Enhanced Routing

One thing our loyal customers have grown to know is that we’ll never stop until we find the “prefect route” between point “A” and “B”.  When we first launched in Singapore, we went through lengths to eliminate the classic routing problem of taking a round-trip through the United States to move your packets across the street.  This involved strategic vendor selections, in-depth discussions with provider engineering staffs about their respective BGP-signaled traffic engineering capabilities, and most significantly, a major push to establish dozens of peering relationships, both at the Equinix Exchange and over private interconnects.  This passion remains strong, and we are now founding members of an initiative to build a new Internet Exchange in Singapore, the SGIX.  In a way, being in Hong Kong is less a major leap for us, and more a small and logical progression towards this goal.

If you’re an Asian carrier, service provider or large enterprise interested in peering with Voxel in either Singapore, Hong Kong or any of our other nearly 20 global locals, please contact peering@voxel.net to discuss.

While many things have changed in the cloud computing arena over the last three years, at the end of the day what you’re buying hasn’t: you’re getting a server with CPU, RAM, and some kind of disk.  Nowhere is this truer than over at Amazon, where you buy CPU power that’s measured in EC2 Compute Units. Amazon defines these as “… equivalent CPU capacity of a 1.0-1.2 GHz 2007 Opteron or 2007 Xeon processor“.

On the surface, it seems quite helpful that they’re abstracting something as complicated as a CPU and offering you a simple to understand unit.  Who wants to look at front side bus speeds?  Who cares where the memory controller lives?  Storage is storage and you buy it in Gigabytes. RAM is RAM, and you buy it in Gigabytes. So why not sell CPUs the same way?  Well, as Intel and AMD will attest, not all 2.4GHz CPUs are created equal.

Suppose you want to buy an EC2 instance with four compute units. Since we’re basing this on 2007 technology, this means that the power might be equivalent to a 2.4GHz AMD Opteron 250 (2×1.2GHz), or it might be equivalent to a 2.0GHz Xeon (2×1GHz). Both of these CPUs offer you two compute units.  In the real world though, the AMD Opteron 250 is more than twice as powerful as the Xeon 2.0 GHz. All of a sudden our yardstick can stretch from two feet to six feet.

The problem with trying to base a unit of CPU on clockspeed is that it’s no longer (and hasn’t been for some time) the sole determining factor in CPU performance. It’s like buying a car based only on horsepower.  On paper, the 403hp Cadillac Escalade is more powerful than a 375hp Ferrari F355 – which do you suppose is faster accelerating to 100mph?

With AgileCLOUD, you’re buying the Ferrari, and it’s made by Intel.  All AgileCLOUD instances are powered by the latest Intel Nehalem CPUs.

Let’s compare an Amazon instance with an AgileCLOUD instance and see what this actually means.  We’ll take a quick look and compare an entry-level AgileCLOUD instance with an Amazon Large instance.

• The Amazon Large Instance is $0.34/hr, and it offers 7.5GB RAM and 4 EC2 Compute units (2 cores, 2 compute units each, meaning 4-4.8GHz of 2007 processing power).
• The AgileCLOUD instance is $0.10/hr, starting with 2GB RAM and a single Nehalem core.

With twice the number of logical cores and twice the clock cycles, you’d probably expect Amazon to come out on top of a multi-threaded Unixbench shootout… you’d be wrong though:

Unixbench Raw Scores (higher is better)

In 10 runs spanning two days, the lower specced AgileCLOUD instance outperformed the EC2 instance in 2 key metrics:

• Raw Performance – On average, the AgileCLOUD instance outperformed EC2 by 7%.
• Consistency – AgileCLOUD’s performance varied by less than 1% through all 12 runs, while EC2 varied by more than 5%. During these random EC2 performance drops, AgileCLOUD pulls ahead by more than 12%.

The most important number of all isn’t found on this graph though, and that’s the cost.  AgileCLOUD outperformed EC2 by an average of 7%, while at the same time coming in at less than one third the cost of the EC2 instance.

Stay tuned, as next week we’ll have a more comprehensive set of benchmarks that take a look at more than just the CPU.