OpenMetal Cloud Customer Success Story
Convesio has pioneered new innovations in the WordPress hosting space since its founding in 2018. Their trajectory, however, was being stifled by outrageous costs, lackluster support, and inflexibility via AWS and GCP. They set out to find cloud providers that they could work closer with to help them continue to revolutionize and grow their platform. OpenMetal helped Convesio gain:
- Control and Customization: Convesio gained root access to their cloud infrastructure to customize for optimal performance.
- Cost Reduction: Convesio has cut their public cloud costs more than 50%.
- A Cloud Partner: Convesio has gained a partner that shares their values and vision.
The Convesio Cloud Wishlist
- Five worker VMs per cluster to host various customer sites and workloads.
- Large VMs to provide burst support.
- Reduce expensive bandwidth costs.
- Reliable primary/failover services.
- More access and control to their supporting infrastructure for:
- Performance optimization.
- Density control.
- Needed assisted management and support for OpenStack.
The Vision
While true fans of WordPress, Convesio had become frustrated by the lack of innovation in the space. The majority of WordPress users still overwhelmingly host their sites on shared servers. It is a low-end, but inexpensive solution for low-traffic sites. Owners of higher-end sites with heavy traffic demands could build and rollout highly scalable infrastructure on public cloud hyperscalers, but this required time and investment to hire developers and engineers. This made scaling feasible for larger sites, but created a gap for mid-market site owners.
To remedy this, mid-market sites began looking to eCommerce platforms like Shopify to solve their scaling challenges. This concerned Convesio, as Shopify is not open source or part of the WordPress space. This trend could mean the loss of high-value companies from those two communities.
Convesio knew they wanted to architect a highly scalable system for WordPress. But what they envisioned was not yet developed in the market. Convesio spent the next three years in research and development working with customers. They tested and validated ways to host WordPress using Docker and microservices to scale containers effectively until they achieved their goal. Finally, with positive feedback and with actual applications in use, they rolled out the service and began selling their product commercially in January of 2020.
With their service rolled out, Convesio immediately jumped into pursuing platform performance improvements by identifying and testing services from different cloud providers.
The Challenge
Convesio’s intention was never to be a one-cloud shop. Their infrastructure was built with the idea of being cloud agnostic. They believe in the benefits of running solutions on multiple clouds for the same site and taking advantage of the ultra redundancy that comes with having a presence with different infrastructure providers.
Convesio deployed services on Amazon Web Services (AWS) and then Google Cloud Platform (GCP). Notably, they broke the norm used by their competitors in choosing not to adopt the internal services, giving them the agility to partner with many different providers.
No Differentiation
Yet, a problem persisted. Convesio realized they were using the same cloud providers as all of their competitors. As a result, they were using the exact same cloud instances and hardware as Kinsta, WP Engine, and a dozen other hosting providers. This made it extremely difficult to differentiate their service.
Convesio knew that to stand out, it would need to deliver exceptional performance. Achieving this, however, required working closely with each cloud provider to customize their hosting services.
This was a concern because Convesio’s experience with AWS and Google was in stark contrast to their own culture and mission to deliver a white glove, hands-on experience for their own customers. They had grown increasingly frustrated by the lack of attention that they received, especially unanswered support requests that would stretch hours and days without a response. Feeling more like an insignificant number than a valued customer, Convesio felt that customization on AWS or Google was not a feasible option.
Needed a Cloud Partner
To overcome this, they set out to find cloud providers that they could work closer with to help them continue to innovate and grow their platform. They expanded their reach to hosting providers one level down, engaging with providers like DigitalOcean, Vultr, and other smaller independent cloud providers. Using these services, Convesio realized they were simply offering virtual private clouds (VPCs) provisioned on a hyperscale provider. Once again, not the innovation they were seeking.
“Your team is the kind of people we would dream of having on our own team. And the great thing about working with you is that it feels like they are part of our team. This quickly won over trust with us.”
Tom Fanelli, CEO and Co-Founder of Convesio
The Solution Journey – Phase One
Consultation and Proof of Concept (PoC) Build
Three XL V2 Private Cloud Cores (Four x 3.2 TB NVMe) with Half Ceph / Half Ephemeral
With agreement on the scope, the team proceeded to build out an initial proof of concept (PoC). This PoC was a development cluster with 60 to 96 vCPU VMs on a XL Cloud Core for testing and learning purposes. It was all built on OpenStack, with half Ceph storage and half Ephemeral storage, as well as integration to a full suite of service capabilities, load balancers, etc. It was a feature rich environment built to give Convesio a complete bench test of the capabilities of OpenMetal.
The PoC was a measured success. Convesio found confidence in the private cloud solution. The hardware infrastructure gave them a path to achieve the levels of customization, innovation, performance, and resiliency they had envisioned. And they found the collaborative partner relationship that they had been looking for. Convesio saw OpenMetal as a partner aligned culturally to serve its customer with the same level of commitment and excellence that Convesio serves its own clients.
The result that sealed the deal for them, however, was the cost. Early indications demonstrated a 50% reduction from the costs they were paying to GCP for each clusters they were deploying.
With a successful PoC and service agreement, OpenMetal and Convesio began building a full production build together.
“It’s really awesome to work with someone who’s aligned culturally to the same type of mission that we are. And it’s really provided us with the ability to innovate and differentiate from the masses that are out there all using the same hyperscalers.”
Tom Fanelli, CEO and Co-Founder of Convesio
The Solution Journey – Phase Two
First Production Build
Three XL V2 Private Cloud Cores (Four x 3.2 TB NVMe) + Two XL V2 Storage and Compute (Four x 3.2 TB NVMe) Half Ceph / Half Ephemeral
To make the investment with OpenMetal financially successful, Convesio needed to be able to deploy a minimum of two large clusters on OpenMetal, similar in capacity to what they had been deploying with GCP, individually. The end goal was to achieve two large clusters protected by triplicate data storage redundancy on Ceph storage.
OpenMetal began by expanding their original PoC design from three to five physical bare metal servers. Convesio wanted to spread out the VMs on their first production cluster, so adding two additional servers allowed for individual VMs on each of the five servers.
It was important to Convesio to deploy large VMs, as they would often have clients that needed resources on standby to allow for bursts of traffic. When bursts happen, they can only burst up to the amount of available resources to the operating system without causing performance slow-downs. Because Convesio was chiefly interested in ensuring performance, this meant that they needed resources available to handle any bursting that may happen.
When you operate in a shared environment, you are still required to pay for these resources that are just sitting idle waiting for the possibility of a burst to happen. As an example, Convesio was paying GCP for approximately 60 vCPUs per VM, when they rarely needed more than five or six vCPUs. But they needed them, just in case they needed to burst. These types of deployments result in wasted resources.
The Need for Bare Metal
Moving to bare metal servers would resolve this. When the Cloud Core servers are idle, Convesio can maximize their existing resources to house more clusters, add more clients, and scale vertically on-demand, without adding additional costs.
The OpenMetal team then reinforced their build footprint with PCLe 4.0 NVMe SSDs to deliver high performance on the latest version of the Ceph storage. The OpenMetal team also helped to customize the size of disks that Convesio was using to maximize space, based on what they thought they needed.
To address early storage issues, OpenMetal helped Convesio enable operating system features such as TRIM to free up space when VM files are deleted. After a number of workload retrospectives, the two teams worked together to determine the optimal ratio of CPU to RAM to DISK, maximizing the density of Convesio’s production build to increase performance and capacity.
“If we hit capacity on disk but still have CPU capacity, we wanted to tap into that. OpenMetal has been extremely collaborative with us to work that out, even building out spreadsheets to model out how many clients and what the density was based on CPU usage and disk usage.”
Tom Fanelli, CEO and Co-Founder of Convesio
The Solution Journey – Phase Three
The Final Production Build
Three XL V2 Private Cloud Cores (Four x 6.4 TB NVMe) + Two XL V2 Storage and Compute (Four x 6.4 TB NVMe) Full Ceph
As Convesio moved into a third phase of their journey, the teams began a second production build. Rather than using the fastest direct NVMe storage, they opted to fully configure the disks for Ceph block storage. Their platform optimization solution already enabled reliable caching for their customers and the direct storage was not adding enough significant boost to performance after many tests. All of the disks were now part of Ceph storage, which gave them even more capacity, while being extremely performant.
Another benefit is that it increased their resiliency (and ability to quickly recover in case of hardware failure, which is not possible with direct storage). OpenMetal also upgraded their hard drives from 3.2 TB to 6.4 TB, making the total Ceph capacity around 116 TBs.
Their original footprint was five server hosts to support a single cluster with five worker VMs per cluster. So they felt that if they could achieve two clusters (double their current footprint) deployed on the proposed XL Cloud Core it would be a successful ROI. The result showed the capability to support between 10 and 12 clusters with XL Cloud Core, significantly exceeding their ROI expectations.
OpenMetal also put Convesio into its upstream bandwidth model that gets billed at the 95th percentile and enables more flat billing rates to prevent unpredictable fees. This allows Convesio more control over data transfers and significantly lowers costs vs the public cloud fees that they were paying to GCP.
Estimated Cost Savings
| Cluster Specifications | Google Cloud Platform (GCP) | OpenMetal |
|---|---|---|
| Cluster Capacity | One Cluster Per | Six+ Clusters Per |
| Size of VMs | Large | Large |
| Number of VMs per Cluster | Five | Five |
| Monthly Pricing Per Cluster: | ||
| Cluster 1 | $3,963 | $9,500 |
| Cluster 2 | $3,963 | $0 |
| Cluster 3 | $3,963 | $0 |
| Cluster 4 | $3,963 | $0 |
| Cluster 5 | $3,963 | $0 |
| Cluster 6 | $3,963 | $0 |
| Total Monthly Cost | $23,778 | $9,500 |
| Bandwidth | Est. $0.02/GB | Est. $0.01/GB |
As shown above, Convesio was purchasing individual GCP clusters with five large VMs per cluster (on average) to support multiple customer sites and workloads. In actuality, not all of the GCP clusters they deployed may have used large VMs. But to simplify this comparison we are comparing their estimated “per cluster” cost of $3,963 on GCP to the cost of using large VMs on their total OpenMetal build at approximately $9,500.
By moving to an OpenMetal Cloud Core with the two additional XL servers, Convesio found that they are able to support somewhere between six to 10 clusters, with each cluster delivering the same VM and resource capacities that they were experiencing from their individual GCP clusters. This exceeded Convesio’s goal of two clusters on OpenMetal by more than three times. But financially, it has allowed Convesio to conservatively deploy six+ clusters on one OpenMetal production build for approximately $9,500 a month versus having to purchase six or individual GCP clusters for $23,778 or more per month.
While the first cluster on OpenMetal is more expensive than the first GCP cluster, it is easy to see how quickly the advantage of adding multiple clusters on OpenMetal adds up to significant savings over the addition of individual clusters on GCP.
When combined with OpenMetal bandwidth costs that are 50% to 80% lower than public cloud costs like GCP, the financial advantages quickly added up.
The Results
Through collaboration, OpenMetal has been able to help Convesio fulfill its goal and create something new and innovative for the WordPress community. But for Convesio, this journey to OpenMetal has also provided them three key advantages:
A Cloud Partner
Convesio has found a team that complements their culture and mission to deliver high touch services and provides ongoing collaboration to build new solutions.
Control and Customization
Convesio now owns their cloud infrastructure, giving them the power to optimize performance and build new innovations for platform and the WordPress community.
Cost Reduction
Convesio has achieved new levels of cloud resource and operational efficiency that have cut their cloud spend more than 50% to use for new business development.
The Next Phase?
For its next phase, Convesio is already working on replicating their current OpenMetal footprint to deploy additional production builds in OpenMetal data center sites that include Los Angeles, California, and Amsterdam in Europe.
“With OpenMetal, we found a true partner, we have more control over the performance of our clouds, and we are able to significantly reduce our cloud costs. These three things make this relationship something I would say yes to a hundred times over.”
Tom Fanelli, CEO and Co-Founder of Convesio
Convesio Recommendation
At the end of this case study, Tom Fanelli, Convesio’s Chief Executive Officer and Co-Founder, was asked what advice he would give to others also considering a move from their public cloud providers. The following is his response.
“Public clouds are really too expensive. You don’t have to spend that level of investment with a public cloud. The answer to that is a private cloud. But you need a trusted expert that you can rely on, and that trusted expert is OpenMetal. With OpenMetal you’re going to have huge savings, but you need an expert to navigate the waters because this technology is very different from the public cloud. So that’s why I would tell people to go with OpenMetal.”
