$400,000,000 Saved - NO MORE AWS

Introduction and Cloud Infrastructure Overview

The speaker discusses the popularity of cloud infrastructure and the advantages it offers, such as flexibility, easy scaling, and instant presence in multiple regions. However, they also mention that the cost of outsourced cloud services can be high.

Cloud Infrastructure Trends

• Cloud infrastructure has become a popular trend in IT.

• Moving to the cloud offers various benefits like flexibility and easy scaling.

• Cloud service providers offer convenient and often user-friendly services.

• AWS's interface is an exception with its outdated design.

Concerns about Cloud Costs

• While clouds provide many benefits, their costs can sometimes be high.

• The speaker expresses concerns about whether their business could survive if all infrastructure was in the cloud.

Comparison of Ref's Own Hardware vs. AWS Cloud

The speaker explains how they conducted a comparison between their own hardware setup in a rented co-location data center in Singapore and a similar installation on Amazon Web Services (AWS) cloud.

Ref's Hardware Overview

• Refs have a rented co-location data center in Singapore with homogeneous infrastructure.

• They consolidated all costs spent on this data center and divided them by the number of servers for comparison.

• Their hardware is relatively new, acquired during the peak of COVID-19 lockdowns in 2020.

• Servers have similar configurations except for CPUs of two generations with the same core count.

• High core count CPUs, terabyte RAM, and 2x 100 gigabits per server are used.

Monthly Cost Calculation

• Refs assumed that all hardware would be used for five years before considering anything beyond that as a freebie.

• The monthly part of initial equipment cost is calculated by dividing the original price by 60 months (5 years).

• Current costs like rent and electricity are as of October 2022, without considering inflation.

Consideration of Electricity Charges and Other Costs

The speaker discusses how they accounted for electricity charges, rental costs, IP transit, dark fiber, and other expenses in their comparison.

Co-location Payments

• Co-location payments include rent and metered electricity.

• Electricity prices have significantly increased since the beginning of 2022.

• Monthly charges were calculated based on the latest higher electricity charges rather than the average for the whole rental period.

Breakdown of Monthly Costs

• Server hardware accounts for two-thirds of the overall monthly expenditure.

• Data center rent with electricity, internet service providers (ISPs) with IP transit and dark fiber, and internal network hardware make up the remaining one-third.

AWS Infrastructure Costs

The speaker mentions that they will discuss AWS infrastructure costs in a future video but does not provide further details in this transcript.

Timestamps beyond this point do not contain relevant information or discussions related to the topic.

EC2 Configuration Comparison

The speaker discusses the comparison between an EC2 configuration and a reference server in terms of cores and RAM. They question whether it is fair to compare the cost of two EC2 instances with the cost of a single reference server.

EC2 Configuration Comparison

• The speaker mentions that they found an EC2 configuration with half the cores and one TB of RAM, which is not as much as their reference server but still considered good.

• They raise the question of whether it is fair to compare the cost of two such EC2 instances with the cost of a single reference server, considering potential discounts for larger machines.

Long-Term EC2 Usage and On-Premise Services

The speaker discusses long-term EC2 usage and compares the lowest price of a three-year reserved EC2 instance with on-premise services. They also mention considerations regarding elastic block storage (EBS) as a replacement for directly attached storage.

Long-Term EC2 Usage and On-Premise Services

• The speaker mentions comparing the lowest price of a three-year reserved EC2 instance with on-premise services for five years.

• They note that they chose the highest price for on-premise services and the lowest price for EC2 instances.

• Elastic Block Storage (EBS) is discussed as not being an accurate replacement for directly attached storage due to differences in speed.

• Cheaper gp3 EBS is chosen, although slower than their current setup.

• Two copies of data chunks are kept on servers, while only usable space in EBS is ordered.

• The price of gp3 storage size is estimated to be equal to half the size of their drivers.

Elastic Block Storage (EBS)

The speaker provides further details about elastic block storage (EBS), including its two-part storage size and charges for input/output operations per second (IOPS). They explain their approach to data replication.

Elastic Block Storage (EBS)

• The speaker explains that EBS consists of two parts: storage size and charges for IOPS.

• They mention using cheaper gp3 EBS, which is slower than their current setup.

• Two copies of data chunks are kept on servers, while only usable space in EBS is ordered.

• The price of gp3 storage size is estimated to be equal to half the size of their drivers.

Limitations of EBS gp3

The speaker discusses limitations of EBS gp3, such as maximum throughput and performance. They compare these limitations with the performance of their current setup.

Limitations of EBS gp3

• The speaker mentions that the maximum throughput for gpt3 instances is 10 gigawatts per second, while a single iacp e Gen4 envoom driver has a performance of six to seven gigawatts per second.

• They highlight that they have 16 drivers working in parallel, leading to significantly higher performance compared to gpt3 instances.

Pricing Considerations and Average Download

The speaker discusses pricing considerations related to AWS charging per gigabyte downloaded. They mention estimating average download per server based on AWS network calculations.

Pricing Considerations and Average Download

• AWS charges per gigabyte downloaded, not for bandwidth.

• The speaker estimates the average download per server and uses this figure in AWS network calculations.

Comparison Analysis

The speaker provides an analysis comparing the costs between co-location and AWS. They discuss factors such as EC2 cost, energy cost, double storage, and consolidation.

Comparison Analysis

• Upon consolidating the analysis, it is found that the imaginary AWS spending is much higher.

• The speaker mentions using the best EC2 cost and worst energy cost for their analysis, giving EC2 an advantage.

• They note that they are not considering double storage in their comparison, only single storage.

• A replacement EC2 instance with similar usable SSD space would cost roughly the same as all 11.3 servers in a co-located data center.

• Accordingly, their 20-server rack would transform into just two servers.

Cost Comparison and Revenue Analysis

The speaker discusses the significant difference in costs between co-location and AWS based on revenue calculations. They consider negotiation possibilities and highlight the high expenses associated with running infrastructure in Singapore.

Cost Comparison and Revenue Analysis

• Even after considering negotiation possibilities, the cost difference between co-location and AWS remains significant.

• The speaker mentions that even if their calculations were off by half, it would still be five times more expensive to use AWS.

• They estimate needing to pay AWS an additional 400 million for infrastructure running in Singapore since 2020.

• The profitability of their company is mentioned but also highlighted as a private company.

The transcript provided does not include further sections or timestamps beyond this point.

New Section

The speaker discusses the use of linear extrapolation for estimating future projections and its limitations.

Linear Extrapolation

• Linear extrapolation is commonly used to estimate future projections.

• However, it tends to over-project given enough time.

• It provides a rough estimate that allows drawing some conclusions.

• The speaker mentions the goal of making more money.

New Section

The speaker discusses the revenue and costs associated with using AWS for data center replacement.

Revenue and Costs

• A ref's total revenue for the last three years would be 257 million.

• However, AWS cost for such a data center replacement would be 448 million.

• The company's revenue wouldn't cover the AWS usage within two and a half years.

New Section

The speaker emphasizes the impact of relying solely on AWS for infrastructure.

Dependency on AWS

• Relying solely on AWS would mean all profits go towards paying for AWS services.

• This scenario raises questions about who benefits from subcontractors' work.

• The speaker refers to a report by Dr. L J Hart Smith that highlights the importance of prime manufacturers or system integrators benefiting from their work.

New Section

The speaker reflects on the advantages already provided to AWS and other costs involved in Aref's operations.

Advantages and Other Costs

• Aref has already provided advantages to AWS, such as higher electricity costs in co-located data centers.

• Only storage parts of EBS were considered, ignoring its slow performance.

• Aref incurs expenses in other areas like additional data centers, services, people, offices, marketing, etc.

New Section

The speaker acknowledges that the article doesn't consider other factors that complicate the comparison between Aref and AWS.

Other Considerations

• The article overlooks aspects like people skills, financial controls, cash flow, capacity planning, and load type.

• These factors would make the comparison between Aref and AWS even more complicated.

New Section

The speaker highlights the significant savings achieved by Aref by not relying 100% on AWS infrastructure.

Infrastructure Savings

• Aref has saved approximately 400 million by ensuring its infrastructure is not solely on AWS.

• This figure continues to grow as they invest in new co-located data centers with their hardware.

• While Aref benefits from hosting their front end on AWS, the majority of their infrastructure is co-located with owned hardware.

• Being entirely dependent on AWS would result in significantly higher costs for Aref.

New Section

The speaker explains how having their own servers allows Aref to achieve faster and better results compared to using cloud services like AWS.

Faster Results with Own Servers

• Aref's servers are much faster than what a cloud service like AWS can provide due to limitations.

• Reports are generated faster and are more comprehensive because the time needed per report is shorter.

• Serverless options may initially be cheaper but can become more expensive as usage increases.

New Section

The speaker discusses the cost considerations of using cloud-only approaches versus owning servers.

Cost Considerations

• Using a cloud-only approach would result in infrastructure costs that are 10 times higher or more compared to owning servers.

• However, there are additional expenses associated with managing employees who maintain these servers.

• Factors such as network engineers, infrastructure engineers, and overall risk need to be considered when evaluating costs.

• The speaker suggests that the cost threshold for considering cloud-only approaches may vary depending on the organization's size and specific circumstances.

The Cost of AWS in Singapore

In this section, the speaker discusses the high costs associated with using AWS in Singapore and the discounts available for startups and hobbyist enterprises.

Retail Pricing of AWS

• The retail pricing of AWS in Singapore is expensive.

• Starting a startup or hobbyist enterprise can be costly.

• However, these types of businesses can get a 50 to 70% discount on AWS services.

Consider Cloud Benefits vs. Actual Cost

• CFOs, CEOs, and business owners interested in sustainable growth should periodically evaluate the benefits of using cloud services like AWS compared to the actual cost.

• While cloud services are often the go-to choice for early-stage startups, as a company grows, maintaining substantial infrastructure on the cloud may become expensive.

Unknown Knowns and Accepting Built-in Ideas

• There are known unknowns and unknown unknowns when it comes to making decisions about cloud infrastructure.

• However, there is also a fourth category: unknown knowns - accepting built-in ideas without realizing why they are accepted.

• It's important to question assumptions and not blindly accept that hosting on AWS or any other cloud provider is the only way to build a company.

Migrating from Cloud Services

This section explores the challenges and considerations involved in migrating from cloud services like AWS to an organization's own infrastructure.

Migrating for Mature Companies

• For more mature companies, migrating from the cloud to their own infrastructure can be difficult.

• It may also be challenging to keep the company alive during such a migration process.

• However, this painful move may ultimately save the company money by avoiding increased revenue cuts to cloud service providers.

Opportunities for Reverse Migration

• Big companies like Fang have dominated the job market by hiring engineers to run their data centers and infrastructure.

• However, recent mass layoffs in big tech companies present an opportunity for smaller companies to reevaluate their approach to the cloud.

• Consideration can be given to a reverse migration from the cloud and hiring seasoned professionals for managing data centers.

Netflix's Open Connect Architecture

This section focuses on Netflix's Open Connect Architecture (OCA) and its role in content delivery.

OCA Network

• Netflix has its own OCA Network, which is an exceptionally sophisticated content delivery network.

• The OCA Network delivers approximately 33% of the internet's content.

• It allows Netflix to efficiently deliver content across the world.

Comparing Data Transfer with YouTube

• Netflix transfers a significant amount of data, possibly more than any other service, including porn websites.

• YouTube, while popular, may not transfer as much data as Netflix.

Evaluating Server Needs for Streaming Videos

This section discusses the server needs for streaming videos and how Netflix manages its network infrastructure.

Netflix's Network Infrastructure

• Netflix has a dedicated network infrastructure called Open Connect that handles video streaming needs.

• The network is designed specifically for delivering high-quality video content efficiently.

Porn vs. Netflix Data Transfer

• While it is believed that porn websites generate significant internet traffic, it is suggested that Netflix may actually transfer more data than them.

• The speaker humorously comments on viewers' addiction to porn but acknowledges that it remains uncertain who transfers more data between porn websites and Netflix.

Please note that these summaries are based solely on the provided transcript and may not capture all the details or nuances of the original video.

New Section

The speaker expresses excitement about an article and asks if everyone is happy with it.

Article Appreciation

• The speaker finds the article to be awesome and great.

• They express their happiness about the article.

• The speaker asks if others are also happy with the article.

No specific timestamps were provided for this section.