This week in our series, Running on OCP, focused on organizations using OCP in various aspects of their infrastructure we speak with one of the earliest adopters of OCP, Actapio, Inc., as well as one of the first OCP Solution Providers, ITOCHU Techno-Solutions America.
Actapio, Inc. is a US subsidiary of Yahoo Japan Corporation and has been at the forefront of OCP technology. ITOCHU Techno-Solutions America is one of our leading Solution Providers and has been working with Actapio on accelerating their OCP infrastructure since the beginning. I was very impressed with Takashi Watanabe, Senior Software Engineer and Server Infrastructure Team Manager at Actapio, Inc., and Hisa Tanaka, Director, Sales & Solution Engineering at ITOCHU Techno-Solutions America, Inc. as they describe their OCP journey and how they have managed a growing infrastructure.
This is a two-part interview with a great deal of insight from an adopter perspective and the importance of a strong partnership.
"My name is Takashi Watanabe. I am a manager and Senior Software Engineer at Actapio, Inc., where I oversee server infrastructure. I started my career at Yahoo Japan Corporation in 2010, working as a server engineer, and my duties expanded to focus on innovation and deployment of Linux as well as hardware platforms including OCP. I also develop software that supports our infrastructure operations. I was assigned to work at Actapio in 2018."
"My name is Hisa Tanaka. I am the Director of Sales and Solution Engineering at ITOCHU Techno-Solutions America which is an OCP Solution Provider (SP) across Asia and the US. I oversee both the business and technical side and I’m based in Silicon Valley. I've been working with the OCP since 2013 and set up the OCP SP relationship with the foundation to enable comprehensive services to new Open Compute adopters. We've been working with Actapio from the beginning of their first adoption of OCP in 2015, and taking this exciting journey to bring their infrastructure to the next level."
"I know that your team has been around OCP since the beginning and it's been great to watch your team grow as Actapio has adopted Open Compute. Many people may not be familiar with your company so Takashi, please share some background on Actapio?"
"Actapio, Inc., formerly known as YJ America, Inc., was founded in 2014 and is a 100% subsidiary of Yahoo Japan Corporation founded in 2014. Actapio’s headquarters is located in central Washington State. Our main service is providing computer resources to Yahoo Japan Corporation as a data center for AI and big data. We currently operate over 7,000 OCP servers, and the number continues to grow. Actapio understands the benefits of being involved in many open communities."
"When you first started your data center in Washington State in 2015 OCP was relatively new. Why did you choose Washington State to build your data center?"
"So, there are three main reasons, the first reason was the low cost of power. Compared with Japan, the electricity cost is much lower in Central Washington because the hydropower industry in that region is able to produce an abundance of clean energy. So we are able to save a lot on power costs.
The second reason is the affordable latency across the US and Japan. The location of the data center is on the west coast, which actually puts us in close proximity to Japan. When we think about the network latency, by looking at the workload of AI and big data, this location is ideal."
Thirdly, having a mature data center ecosystem in the region has allowed us to benefit from best practices of other companies. For example, when Yahoo Japan Corporation builds a data center, they must take into account the fact that earthquakes can have a significant impact on the construction schedules. Actapio benefited from other hyperscale data centers in the area developing partnerships through open communities with companies like Facebook, Google and Microsoft. One additional, and unexpected benefit from building our data center in Washington State is that we are in the same time zone with many other hyperscale companies and startups. This is huge benefit because it allows us to more easily have discussions with other local hyperscalers, as well as startups in the region and on the west coast of the US (i.e. Silicon Valley)."
"That's interesting about the time zones and being able to work together more closely with teams in the bay area.
Now, when you first started looking at Open Compute designs, what were your expectations and what were some of the biggest challenges you faced back in 2015?"
"A few challenges led us to look at OCP as the right strategy. Our biggest challenge initially was limited resources. Actapio had only a few engineers when it was founded back in 2014, so being able to service a large number of servers with limited engineers was a key to success. Since Actapio has fewer human resources than Yahoo Japan Corporation, the number of servers a single technician or engineer can service was crucial. When we started to talk with the Open Compute Community members like Facebook and others, OCP was already proving to be an infrastructure that is possible to operate and service at scale.
Secondly, there was also interest in working on common domains and common opportunities. For example, Actapio was pursuing big data infrastructure so it is a common area that we can benefit from by working with others in the Community such as Facebook.
Thirdly, by working in an open community we are able to avoid potential risks by leveraging designs that have been shared and vetted by the vast community of engineers from a variety of companies. So that was very important in Actapio’s decision to look at Open Compute as the right path to pursue."
"Were there any challenges or expectations that were more significant than normal when you deployed the hardware?"
"I think the biggest expectation that Actapio had for OCP was operational efficiency. In order to allow for few engineers to operate the larger infrastructure, that was the best choice, but there were certain challenges."
"First of all, we had to build the partnerships in order to build a supply chain. So we built a relationship with ODM’s, key parts manufacturers and integrators to bring the full OCP solution to the data center.
Then, a second challenge was standardizing Quality Assurance (QA) items and processes across multiple vendors. When you build something at scale, you will not rely on a single vendor. So we pursued a strategy to do dual sourcing which allowed us to choose a single design and because the OCP design is open source it's pretty much the same design. But we had to choose at least two vendors to build our OCP supply chain. In order to ensure we met Actapio’s quality standards, we had to create QA items and processes to be implemented across multiple vendors. We have continued to refine, improve and automate this over the last five years, but this first step was quite a challenge.
The third challenge, which is not as significant today because we have optimized our procurement, but initially we had to accommodate a long lead time. So basically, OCP is built to order (BTO) so we had to establish solid communications with our ODM partners and share our implementation plan ahead of schedule to ensure a proper forecast of our infrastructure needs.
In the legacy way of procuring infrastructure we would have a basic specification, desired quantity, and cost along with the standard delivery time. With OCP’s open architecture we can pick and choose the key components which work best. We basically create a recipe from the beginning. So, while we benefited from the flexibility, the initial side effect of BTO was an area we had to address through developing solid communication channels with our supply chain partners early in the process.
The last challenge was the validation process. OCP allows us to have a choice of a motherboard, CPU, memory, SSDs, NIC and most everything. We needed to make sure these components are well validated with bios, BMC's components and configurations. So Actapio and ITOCHU Techno-Solutions America partnered together to enable this validation process. Over time, we have been building expertise and knowledge on how to do this validation process in a more standardized way. The beginning was very hard because all the components need to be validated. It was something that we built together and continues to get better as we expand our infrastructure."
"The QA process, supply chain, forecasting lead time and validation process are all very big challenges which you’ve seem to overcome.
What benefits have you seen in the last 5 to 7 years using Open Compute? Have you seen a reduction in OpEx and CapEx?"
"We would really like to emphasize that a reduction of OpEx was the most important benefit. For example, with OCP we can maintain the server without the need for any tools, so that tool-less maintenance is allowing Actapio’s resources to be more efficiently allocated in the data center. Especially for those components that usually fail at a high frequency, such as DRAM, or other components that have fairly higher failure rate. Once you’ve experienced OCP’s tool-less maintenance and mechanical design, servicing a regular 19 inch traditional server requires a lot more labor. OCP saves a lot of time.
OCP racks also have front serviceability which means you service the rack from the front and not the rear of the rack. Technicians and engineers can stay in a single aisle to service the racks instead of spending time moving between the front and rear of the racks. This is important when you operate at scale with hot aisle and cold aisle containment. It keeps your technicians from excessive exposure to the hot aisle (rear). So it’s beneficial for your employees’ health if they are not exposed to varying temperatures in the data center on a daily basis.
The third reason is the flexibility of the choice of the parts. We’ve been standardizing key components such as direct SSD, NIC, and other components within a server. They allow us to pick, choose and determine the best possible candidates yielding a higher ROI. This is very effective because if we didn't have a choice of the parts and had to compromise we might have piles of spares from different vendors which decreases our operational efficiency. In summary the benefits are ease of maintenance and access as well as the flexibility of choosing the right components."
Part 1 of this interview with Takashi and Hisa highlighted how and why they started to evaluate OCP as a new way of deploying infrastructure. We also covered how they overcame certain challenges in the supply chain and the benefits realized.
In Part 2 of this interview, to be released next week, we discuss more about their involvement in open source, Actapio’s move toward AI and the importance of a strong partnership with ITOCHU Techno-Solutions America. Part 2 will be published next week on our OCP blog page and across our social media outlets.
In the meantime, if you would like to learn more about Actapio’s OCP infrastructure please contact me and I would be happy to introduce you. You can also view all the OCP related products offered by ITOCHU Techno-Solutions America on our OCP Marketplace here as well as fully integrated solutions based on OCP here.
Takashi Watanabe - Senior Software Engineer and Server Infrastructure Team Manager at Actapio, Inc.
Takashi has over 10 years of experience in the IT industry. He has been responsible for the design, deployment, and operation of large-scale infrastructure as a server engineer, and continues to develop software for maximizing IT infrastructure efficiency. He is a proponent for the use and adoption of open-source hardware and software, and holds a Master in Computer Science.
Hisa Tanaka - Director, Sales & Solution Engineering at ITOCHU Techno-Solutions America, Inc.
Overseeing the US business growth out of Silicon Valley. Hisa spent the first 7 years of his career as SE for Telco to build the mobile Internet in Japan. After landing in the US, he has worked with many technology startups to bring innovation to reality, and a part of that was to set up an OCP SP partnership and the success with Actapio. He is a proponent of open technologies and believes operating expertise pioneered by internet giants, but translated to solve existing enterprise challenges, will bring industry to the next level.