NTT Multimedia Communications Laboratories, Inc.

Comments(author unknown)069412432183832489270694639331581020230404730030429766811798

A couple of weeks back Greg Linden sent me an interesting paper called Energy Proportional Datacenter Networks. The principal of energy proportionality was first coined by Luiz Barroso and Urs Hölzle in an excellent paper titled The Case for Energy-Proportional Computing. The core principal behind energy proportionality is that computing equipment should consume power in proportion to their utilization level. For example, a computing component that consumes N watts at full load, should consume X/100*N Watts when running at X% load. This may seem like a obviously important concept but, when the idea was first proposed back in 2007, it was not uncommon for a server running at 0% load to be consuming 80% of full load power. Even today, you can occasionally find servers that poor. The incredibly difficulty of maintaining near 100% server utilization makes energy proportionality a particularly important concept.

                                                                                                                         

One of the wonderful aspects of our industry is, when an attribute becomes important, we focus on it. Power consumption as a whole has become important and, as a consequence, average full load server power has been declining for the last couple of years. It is now easy to buy a commodity server under 150W. And, with energy proportionality now getting industry attention, this metric is also improving fast. Today a small percentage of the server market has just barely cross the 50% power proportionality line which is to say, if they are idle, they are consuming less than 50% of the full load power.

 

This is great progress that we’re all happy to see. However, it’s very clear we are never going to achieve 100% power proportionality so raising server utilization will remain  the most important lever we have in reducing wasted power. This is one of the key advantages of cloud computing. When a large number of workloads are brought together with non-correlated utilization peaks and valleys, the overall peak to trough load ratio is dramatically flattened. Looked at it simplistically, aggregating workloads with dissimilar peak resource consumption levels tends to reduce the peak to trough ratios. There need to be resources available for peak resource consumption but the utilization level goes up and down with workload so reducing the difference between peak and trough aggregate load levels, cuts costs, save energy, and is better for the environment.

 

Understanding the importance of power proportionality, it’s natural to be excited by the Energy Proportional Datacenter Networks paper. In this paper, the authors observer “if the system is 15% utilized (servers and network) and the servers are fully power-proportional, the network will consume nearly 50% of the overall power.” Without careful reading, this could lead one to believe that network power consumption was a significant industry problem and immediate action at top priority was needed. But the statement has two problems. The first is the assumption that “full  energy proportionality” is even possible. There will always be some overhead in having a server running. And we are currently so distant from this 100% proportional goal, that any conclusion that follows from this assumption is unrealistic and potentially mis-leading.

 

The second issue is perhaps more important: the entire data center might be running at 15% utilization. 15% utilization means that all the energy (and capital) that went into the datacenter power distribution system, all the mechanical systems, the servers themselves, is only being 15% utilized. The power consumption is actually a tiny part of the problem.  The real problem is the utilization level means that most resources in a nearly $100M investment are being wasted by low utilization levels.There are many poorly utilized data centers running at 15% or worse utilization but I argue the solution to this problem is to increase utilization. Power proportionality is very important and many of us are working hard to find ways to improve power proportionality. But, power proportionality won’t reduce the importance of ensuring that datacenter resources are near fully utilized.

 

Just as power proportionality will never be fully realized, 100% utilization will never be realized either so clearly we need to do both. However, it’s important to remember that the gain from increasing utilization by 10% is far more than the possible gain to be realized by improving power proportionality by 10%. Both are important but utilization is by far the strongest lever in reducing cost and the impact of computing on the environment.

 

Returning to the negative impact of networking gear on overall datacenter power consumption, let’s look more closely. It’s easy to get upset when you look at net gear power consumption. It is prodigious. For example a Juniper EX8200 consumes nearly 10Kw. That’s roughly as much power consumption as an entire rack of servers (server rack powers range greatly but 8 to 10kW is pretty common these days). A fully configured Cisco Nexus 7000 requires 8 full 5kW circuits to be provisioned. That’s 40kW or roughly as much power provisioned to a single network device as would be required by 4 average racks of servers. These numbers are incredibly large individually but it’s important to remember that there really isn’t all that much networking gear in a datacenter relative to the number of servers.

 

To get precise, let’s build a model of the power required in a large datacenter to understand the details of networking gear power consumption relative to the rest of the facility. In this model, we’re going to build a medium to large sized datacenter with 45,000 servers. Let’s assume these servers are operating at an industry leading 50% power proportionality and consume only 150W each at full load. This facility has a PUE of 1.5 which is to say that for every watt of power delivered to IT equipment (servers, storage and networking equipment), there is ½ watt lost in power distribution and powering mechanical systems. PUEs as high as 1.2 are possible but rare and PUEs as poor as 3.0 are possible and unfortunately quite common. The industry average is currently 2.0 which says that in an average datacenter, for every watt delivered to the IT load, 1 watt is spent on overhead (power distribution, cooling, lighting, etc.).

 

For networking, we’ll first build a conventional, over-subscribed, hierarchical network.  In this design, we’ll use Cisco 3560 as a top of rack (TOR) switch. We’ll connect these TORs 2-way redundant to Juniper Ex8216 at the aggregation layer. We’ll connect the aggregation layer to Cisco 6509E in the core were we need only 1 device for a network of this size but we use 2 for redundancy. We also have two border devices in the model:

 

Using these numbers as input we get the following:

 

·         Total DC Power:                                10.74MW (1.5 * IT load)

·         IT Load:                                                7.16MW (netgear + servers and storage)

o   Servers & Storage:          6.75MW (45,100 * 150W)

o   Networking gear:             0.41 MW (from table above)

 

This would have the networking gear as consuming only 3.8% of the overall power consumed by the facility (0.41/10.74). If we were running at 0% utilization which I truly hope is far worse that anyone’s worst case today, what networking consumption would we see? Using the numbers above with the servers at 50% power proportionally (unusually good), we would have the networking gear at 7.2% of overall facility power (0.41/((6.75*0.5+0.41)*1.5)). 

 

This data argues strongly that networking is not the biggest problem or anywhere close. We like improvements wherever we can get them and so I’ll never walk away from a cost effective networking power solution. But, networking power consumption is not even close to our biggest problem so we should not get distracted.

 

What if we built an modern fat tree network using commodity high-radix networking gear along the lines alluded to in Data Center Networks are in my Way and covered in detail in the VL2 and PortLand papers? Using 48 port network devices we would need 1875 switches in the first tier, 79 in the next, then 4, and 1 at the root. Let’s use 4 at the root to get some additional redundancy which would put the switch count at 1,962 devices. Each network device dissipates roughly 150W and driving each of the 48 transceivers requires roughly 5w each  (a rapidly declining number). This totals to 390W per 48 port device. Any I’ve seen are better but let’s use these numbers to ensure we are seeing the network in its worst likely form. Using these data we get:

 

·         Total DC Power:                                11.28MW

·         IT Load:                                                7.52MW

o   Servers & Storage:          6.75MW

o   Networking gear:             0.77MW

 

Even assuming very high network power dissipation rates with 10GigE to every host in the entire datacenter with a constant bisection bandwidth network topology that requires a very large number of devices, we still only have the network at 6.8% of the overall data center. If we assume the very worst case available today where we have 0% utilization with 50% power proportional servers, we get 12.4% power consumed in the network (0.77/((6.75*0.5+0.77)*1.5)).

 

12% is clearly enough to worry about but, in order for the network to be that high, we had to be running at 0% utilization which is to say, that all the resources in the entire data center are being wasted. 0% utilization means we are wasting 100% of the servers, all the power distribution, all the mechanical systems, and the entire networking system, etc. At 0% utilization, it’s not the network that is the problem. It’s the server utilization that needs attention.

 

Note that in the above model more than 60% of the power consumed by the networking devices were the per-port transceivers. We used 5W/port for these but overall transceiver power is expected to drop down to 3W or less over the next couple of years so we should expect a drop in network power consumption of 30 to 40% in the very near future.

 

Summarizing the findings: My take from this data is it’s a rare datacenter where more than 10% of power is going to networking devices and most datacenters will be under 5%. Power proportionality in the network is of some value but improving server utilization is a far more powerful lever. In fact a good argument can be made to spend more on networking gear and networking power if you can use that investment to drive up server utilization by reducing constraints on workload placement.

 

                                                                                --jrh

 

James Hamilton

e: jrh@mvdirona.com

w: http://www.mvdirona.com

b: http://blog.mvdirona.com / http://perspectives.mvdirona.com

 

From Perspectives.James Hamilton0579891984749163291211295174769146265207024409950815353401250684032643414530167014237529447171936435160086028796463282500298867174475874805011183392210543887492

This week brought news that pharmaceutical giant Eli Lilly has ended its use of Amazon EC2 because of an inability to negotiate contractual liability with Amazon Web Services in the case of outages or data breaches. Though subsequently retracted in this particular case, these negotiations will become more common as large enterprises think about moving more workloads to the cloud. However, as I explain in my weekly column at GigaOM Pro, what’s not so certain is whether cloud computing providers will have to budge.

Cloud computing terms of service — not just from AWS, but from every provider — uniformly deny all liability for outages or data losses, disclaim all warranties of any type, and limit damages to those outlined in the SLA. Although these terms might sound unfair, courts have, in similar circumstances, (i.e., Google AdWords and clickwrap contracts) been quite willing to let those terms stand as they are. This is especially true in cases like that of AWS and Eli Lilly, where both parties are commercial entities perfectly capable of understanding contractual terms and obligating themselves however they please.

When viewed in light of the cloud computing model, there’s even better reason to see why such terms are allowed to persist. The multitenant and anonymous natures of cloud offerings like Amazon Web Services mean that a single failure could result in untold numbers of customers filing lawsuits, even if the cloud provider might not have known they were customers in the first place. That’s a lot of risk for a low-margin business like selling bare VMs.

If self-service cloud providers, such as AWS, really want enterprise customers, they must consider how far they’re willing to bend to earn that business. In some cases, negotiation might be an option. Enterprises, in turn, need to figure out what they actually want from the cloud, because with most cloud providers, they can’t have their cake and eat it too. There are plenty of cloud options offering negotiable contracts, meaningful SLAs and even dedicated resources, but they don’t accept American Express.

Read the full post here.

This week brought news that pharmaceutical giant Eli Lilly has ended its use of Amazon EC2 because of an inability to negotiate contractual liability with Amazon Web Services in the case of outages or data breaches. Though subsequently retracted in this particular case, these negotiations will become more common as large enterprises think about moving more workloads to the cloud. However, as I explain in my weekly column at GigaOM Pro, what’s not so certain is whether cloud computing providers will have to budge.

Cloud computing terms of service — not just from AWS, but from every provider — uniformly deny all liability for outages or data losses, disclaim all warranties of any type, and limit damages to those outlined in the SLA. Although these terms might sound unfair, courts have, in similar circumstances, (i.e., Google AdWords and clickwrap contracts) been quite willing to let those terms stand as they are. This is especially true in cases like that of AWS and Eli Lilly, where both parties are commercial entities perfectly capable of understanding contractual terms and obligating themselves however they please.

When viewed in light of the cloud computing model, there’s even better reason to see why such terms are allowed to persist. The multitenant and anonymous natures of cloud offerings like Amazon Web Services mean that a single failure could result in untold numbers of customers filing lawsuits, even if the cloud provider might not have known they were customers in the first place. That’s a lot of risk for a low-margin business like selling bare VMs.

If self-service cloud providers, such as AWS, really want enterprise customers, they must consider how far they’re willing to bend to earn that business. In some cases, negotiation might be an option. Enterprises, in turn, need to figure out what they actually want from the cloud, because with most cloud providers, they can’t have their cake and eat it too. There are plenty of cloud options offering negotiable contracts, meaningful SLAs and even dedicated resources, but they don’t accept American Express.

Read the full post here.

Social-game maker Zynga has signed a joint venture deal with Softbank Corp. to set up a Japanese subsidiary of the company. Softbank has also invested $150 million in Zynga, whose Farmville and Mafia Wars games are among the most popular on Facebook. The partnership was reported last month by the Nikkei news service and by Bloomberg, but was not officially confirmed until Wednesday. The deal is expected to involve distributing Zynga games to Softbank’s mobile phone customers, one of the largest groups in the Japanese mobile market.

Zynga is also reportedly involved in a partnership with Google that included $100 million in financing from the search company. Although there has been no official confirmation of the investment, Google CEO Eric Schmidt told the New York Times in an interview that a partnership with Zynga was in the works. The web giant is said to be working on a gaming-centered social networking project — code-named Google Me — that is seen as a competitive response to the growth of Facebook.

Zynga has been the social-gaming industry leader for some time, with several of the leading games on Facebook, and has attracted more than $500 million in investment from venture funds (if the reported Google investment is included). Although the company has been closely aligned with Facebook, it has also signed a number of external development and distribution deals, including a recent one with Yahoo that will see Zynga games appear on Yahoo properties. Based on the trading of Zynga stock on secondary private markets, some have estimated the company could be worth as much as $5 billion.

Zynga competitor Playdom was acquired this week by Walt Disney Co. for as much as $763 million, depending on future performance, and game-maker Electronic Arts bought Playfish for as much as $400 million last year.

Related content from GigaOM Pro (sub req’d): How the Next Zynga Could Reinvent Social Gaming

Given that Asia dominates the list of 100 Fastest Internet cities and China is the most populous nation in the world, it shouldn’t come as a surprise to anyone that China is home to the largest broadband service provider in the world. The latest data gathered by Telegeography, a research firm, shows that at the end of the first quarter of 2010, China Telecom led the top 10 broadband service providers rankings. It was followed by China Unicom.

The two Chinese ISPs account for nearly 20 percent of the world’s broadband subscribers. At the end of the first quarter of 2010, there were close to 492 million broadband subscribers worldwide, Telegeography notes. (Related: there are nearly half a billion broadband subscribers worldwide.)

China vs US

The rise of these two carriers also mirrors the rise of China as an Internet behemoth, pushing the U.S. into the second spot. The United States had four broadband ISPs in the top ten list: Comcast, AT&T, Time Warner Cable (c TWC) and Verizon. Korea Telecom cracked the top 10 and edged out Telecom Italia, pushing it to the No. 11 spot.

These top 10 broadband service providers in total account for roughly 39 percent of world’s total broadband customers, and they collectively added about 23.3 million new subscribers between the first quarter of 2009 and the first quarter of 2010.

Who’s The Best Performing ISP?

While sheer size and scope matters, it’s also important to check out a service provider’s actual performance. Many ISPs make wild bandwidth and speed claims, but deliver a suboptimal performance from their broadband connections.

Ookla, the Seattle, Wash.-based company behind the popular Speedtest.net web service, has developed a new methodology (announced yesterday) that shows you the actual performance of broadband service providers from an end-user’s perspective.

“The new ISP ranking data takes a giant step in that direction, further empowering consumers for the first time with rich data that helps evaluate ISP performance close to home or throughout the world,” Ookla co-founder Mike Apgar said in a statement.  With nearly a million speed tests a day, Ookla has a good idea about which ISPs are performing well and are true to their claims. According to Ookla, U.S. ISPs deliver an average download speed of 9.9  Mbps while Chinese ISPs deliver an average download speed of 3.5 Mbps. South Korean download speeds, meanwhile, are around 31.5 Mbps.

How do U.S. broadband service providers rank? Ookla says Comcast is the best broadband provider in download speed terms, with an average download speed of 16.26 Mbps, while Time Warner Cable has an average speed of 13.48 Mbps and Verizon has an average of 10.76 Mbps.

Here are the top ten broadband service providers in the US based on download speeds.

1. Comcast 16.26 Mbps

2. Charter Communications 15.06 Mbps

3. Midcontinent Communications 14.36 Mbps

4. Optimum Online (Cablevision) 14.31 Mbp

5. RoadRunner (Time Warner Cable) 13.48 Mbps.

6. Cox Communications 13.41 Mbps

7. Insight Communications 11.75 Mbps

8. Surewest Broadband 11.44 Mbps

9. Verizon 10.76 Mbps

10. Prairiewave Telecommunications 10.52 Mbps.

You can see clearly that cable companies dominate this list.

Related Research Report from GigaOM Pro: The Internet of Things: Anywhere, Anytime, Anything. Buy for $199 now.

According to Ookla, in the U.S., the average monthly cost for broadband is at $47.32 or about $5.06 per megabit per second. In California, broadband costs just $4.24 per Mbps, while residents of Idaho pay $8.80 per Mbps, Washington respondents averaged just $3.89 per Mbps and Michigan subscribers pay $6.36.

Broadband Image courtesy of Flickr user Gavin St. Ours

Comments(author unknown)

Comments(author unknown)

Comments(author unknown)0689410808472502418505378525178073980996

Comments(author unknown)16035476708569753200

Comments(author unknown)16462790866243475283

There is a resurgence of activity taking place around virtual desktops — where enterprises take their desktop compute environments, and make them configurable, deployable and manageable from a central location. The idea has been hanging around the fringes of IT for years, but I think the time may be right for businesses to actually deploy virtual desktops, for a variety of reasons. And since I think the time is right, I’ve also taken a look at some of the smaller companies that will challenge Citrix and VMware in this emerging sector.

The time is now

Virtualization is better understood and more widely implemented every day. The low-hanging fruit was in the data center, where the impact was less direct to end users and more about cutting excessive server spending and driving up overall server utilization. Now it’s time to find more opportunities to cut costs through virtualization elsewhere in the enterprise.

Devices, operating systems and connectivity are proliferating. Remember when you just had a desktop and a mobile phone that only made calls? For many, those days are long gone, and companies find themselves having to support desktops, laptops and mobile devices across a range of operating systems and a widely dispersed geographic reach. Getting away from hardware dependence and moving to a software-based virtual desktop model simplifies maintenance and reduces costs, helping companies handle device proliferation.

Enterprises are always looking at ways to reduce costs. Virtual desktops promise a host of management and infrastructure savings, including simpler deployment, less storage, and more robust security and compliance. For example, when many similar desktops are created, there is a huge opportunity to consolidate those common operating system images through de-duplication.

Hosted Virtual Desktops promise a massive buying change for enterprise infrastructure. Gartner estimates that the worldwide hosted virtual desktop (HVD) market will accelerate through 2013 to reach 49 million units, up from more than 500,000 units in 2009. Imagine the opportunity for a large company to have the entire desktop infrastructure for thousands of employees hosted externally as a service. The slashing of capital expenses and migration to a subscription-based model is compelling for most businesses.

Who might win?

Many large technology companies have significant product offerings targeting this space, in particular Citrix, Microsoft, VMware, NetApp, Cisco and others. But there’s plenty of interesting startup activity to follow, as well. So, with that in mind, let’s see who’s making waves around virtual desktop infrastructure.

Atlantis Computing offers its ILIO appliance, which it claims can shrink the storage footprint for virtual desktops by 20 times while increasing performance by a factor of 10. The company has an extensive partner list and does not attempt to offer a complete VDI solution itself.

Desktone offers the Virtual-D platform that allows enterprises to offer virtual desktops internally, as well as for service providers to host desktops off-site. The Desktone Access Fabric links these two worlds when needed. Desktone counts Citrix as an investor.

Viewfinity provides systems management, priviledge management, and user migration using an underlying virtualization technology that encapsulates existing OSes and applications to facilitate these functions. The company needs to tighten its messaging around virtual desktops.

Virtual Computer claims that its NxTop product is the industry’s first “bare metal” client hypervisor that provides true isolation and great performance, two issues that can be tricky for enterprises adopting virtual desktops.

Neocleus also uses a bare metal hypervisor to create separate instances of Windows on a single machine so applications can be contained to a single operating systems. The company did a partnership deal with Big Fix earlier this year, and Big Fix was acquired by IBM in June.

Unidesk helps customers manage virtual desktops in VMware and Citrix environments. The company’s Composite Virtualization technology separates the operating system from the application and user personalization layers for easier management. Unlike many others in this roundup, Unidesk lists a few customers on its website.

Leostream has a Connection Broker product that leverages the infrastructure of virtualization providers like VMware, Citrix, Microsoft and IBM. Delivered as a virtual appliance, it manages the connection between end users and their virtual desktops and applications. Leostream also has a long customer list on its site.

Pano Logic offers a range of products and software tools that it bundles within the Pano System, including the Pano Device, a small network-connected “zero client” that has no CPU or operating system and simply connects to a Windows desktop image on a central server.

MokaFive which was founded in 2005 with a lot of noise around the concept of a LivePC has now focused squarely on the virtual desktop market. LivePCs are hosted locally, though they can be created, deployed and managed from a central location.

RingCube offers their vDesk solution which creates personalized workspaces. These workspaces can be hosted directly on a PC, a USB drive, a network drive, or through a virtual desktop hosted on a set of virtualized servers.

With enterprises looking to reign in the complexity and cost of managing employee computer workspaces, and the availability of on-premise and cloud solutions, it is only a matter of time before a few strong leaders emerge around virtual desktop infrastructure.

Gary Orenstein is the host of The Cloud Computing Show.

This week, a relatively large group of technology companies, along with NASA, launched OpenStack, an open source project designed to give businesses and service providers a top-to-bottom, and already proven, cloud computing platform. I’m all for openness, but as I discuss in my weekly column at GigaOM Pro, it’s do not too difficult to play devil’s advocate and make the case that OpenStack won’t create true rivals for leading cloud providers or cloud software vendors.

Assuming the provisioning engine comes to fruition, OpenStack will undoubtedly see adoption from service providers wanting to offer cloud computing, enterprises wanting to build their own internal clouds, and IT vendors looking to beef up their cloud software offerings. If all comes together as planned, it should be a very nice solution. Just like everywhere else in life, competition in the cloud is a great thing.

However, competition for dollars and developers is plentiful, including large cloud providers, software vendors and even other open source options. The idea of a network of interoperable OpenStack-based public and internal clouds is appealing, but it would require stealing business and developers from a wide variety of other proven, innovative and commercially supported offerings.

Even within the OpenStack membership, conflicts of interest exist. How supportive can Rackspace really be of other service providers without cannibalizing its own cloud revenue? Plus, it’s also considering offering Windows Azure as a service. Dell already sells Joyent’s cloud software, and it’s also an early Windows Azure Appliance partner. If Dell does offer OpenStack as an open source alternative, it will be just that — an alternative. Other partners will support the OpenStack platform, but that’s on top of existing support for AWS, VMware and other industry-leading offerings. Supporting OpenStack is one thing, but pushing it to the exclusion of other options is something else.

Is OpenStack important? Yes. Will OpenStack attract a broad community of users? Yes. Will OpenStack-based offerings and deployments gather enough market share to make current leaders lose sleep? That’s not such an obvious answer.

Read the entire column here.

Photo courtesy of Picasa user elinenberg.

Infoworld: "VMware's hosted or Type-2 hypervisor has been around for nearly a decade. But has this virtualization platform reached end of life?"(author unknown)

Comments(author unknown)15932500879777387177

Comments(author unknown)0231204291182305909818302512749180636793023722656982926064730872565204116737808700762146315878389518

Comments(author unknown)

Twitter has decided to put the millions it has raised toward building its own data center, according to a blog post published this evening by the social messaging service. The data center, which will be located around Salt Lake City, Utah, should come online in the fall. The company also says it plans to “bring additional Twitter managed data centers online over the next 24 months.”

We wondered if Twitter was going to build its own data center back in April after John Adams, a Twitter engineer, mentioned the proposition on a slide. However, I ultimately thought it wouldn’t, given that it had just signed a deal with NTT America to expand its dedicated hosting space. Apparently the deal with NTT stays, but the demands of keeping up with the 300,000 people who sign up to the service on an average day requires more customization and more control, which is the main reason Twitter gives for this decision. It also wrote:

Second, Twitter will have full control over network and systems configuration, with a much larger footprint in a building designed specifically around our unique power and cooling needs. Twitter will be able to define and manage to a finer grained SLA on the service as we are managing and monitoring at all layers. The data center will house a mixed-vendor environment for servers running open source OS and applications. Third, having our own data center will give us the flexibility to more quickly make adjustments as our infrastructure needs change.

Indeed, having your own data center is a big milestone for large-scale web services. Facebook announced a data center in January, and despite the lure of the cloud, other webscale operations are keeping some of their servers. Twitter had started out in the cloud, but moved from Amazon’s EC2 to Joyent and then to dedicated NTT hosting, after deciding latency in the cloud was too high. Perhaps Twitter, like Facebook and Google (the granddaddy of webscale infrastructure), has decided that aside from custom code, it needs a custom home for its hardware to ensure a strategic advantage.

Related GigaOM Pro research (sub req’d):

• For Open Cloud Computing, Look Inside Your Data Center
• Social Networks Need to Grin and Bear Infrastructure Costs

Twitter has decided to put the millions it has raised toward building its own data center, according to a blog post published this evening by the social messaging service. The data center, which will be located around Salt Lake City, Utah, should come online in the fall. The company also says it plans to “bring additional Twitter managed data centers online over the next 24 months.”

We wondered if Twitter was going to build its own data center back in April after John Adams, a Twitter engineer, mentioned the proposition on a slide. However, I ultimately thought it wouldn’t, given that it had just signed a deal with NTT America to expand its dedicated hosting space. Apparently the deal with NTT stays, but the demands of keeping up with the 300,000 people who sign up to the service on an average day requires more customization and more control, which is the main reason Twitter gives for this decision. It also wrote:

Second, Twitter will have full control over network and systems configuration, with a much larger footprint in a building designed specifically around our unique power and cooling needs. Twitter will be able to define and manage to a finer grained SLA on the service as we are managing and monitoring at all layers. The data center will house a mixed-vendor environment for servers running open source OS and applications. Third, having our own data center will give us the flexibility to more quickly make adjustments as our infrastructure needs change.

Indeed, having your own data center is a big milestone for large-scale web services. Facebook announced a data center in January, and despite the lure of the cloud, other webscale operations are keeping some of their servers. Twitter had started out in the cloud, but moved from Amazon’s EC2 to Joyent and then to dedicated NTT hosting, after deciding latency in the cloud was too high. Perhaps Twitter, like Facebook and Google (the granddaddy of webscale infrastructure), has decided that aside from custom code, it needs a custom home for its hardware to ensure a strategic advantage.

Related GigaOM Pro research (sub req’d):

• For Open Cloud Computing, Look Inside Your Data Center
• Social Networks Need to Grin and Bear Infrastructure Costs

Comments(author unknown)