Freedom from downtime and latency with our dedicated bare metal servers.
Handle heavy real-time workloads with unparalleled speed and performance.
Bare metal performance with the flexibility of the cloud.
Effective server-side and tech-agnostic cheat detection.
Scaling game instances with automated global orchestration.
Low-latency game server management for a flawless experience.
Custom tools to detect, intercept, and deflect impending attacks.
Transfer data on a global scale fast, private, and secure.
Reach eyeball networks through meaningful peering relationships.
Go global with our custom and secure privately-owned data center.
Discord has about 180 employees, from which a small team of four engineers specifically work on Discords voice and video service infrastructure. The main motivation to create Discord came from the need to make services that use real-time communication (WebRTC), like TeamSpeak and Skype, better. Mark Smith, Director of infrastructure is responsible for everything Discord does at the back end, ranging from cloud systems to voice and video systems, the software that Discord builds, and the hardware they use. He explains the challenges Discord had to face as a start-up:
“We wanted to be the best real-time communication app, but to do that, we had to be able to deliver low latency and high-quality audio services around the world. Since we were a start-up, we were not yet equipped to build up everything that is needed to develop this entire platform around the app by ourselves. We did not have the talent that understood how to build a network or how to maintain the hardware and power in data centers. In addition, we did not have the physical resources around the globe to make this happen. Besides, back in the days as a small start-up, we coped with a challenging timeframe, therefore we needed to find partners to make it happen.”
From the start, Funcom’s engineering approach was designed to eliminate guesswork at scale. Rather than rely on abstracted cloud platforms, they opted for a proprietary K8s-based orchestration layer running directly on bare metal. Each server plays a well-defined role: whether part of a multi node battlegroup, a Kubernetes control plane, or a game logic container.
To support this architecture, Funcom standardized their global fleet on a single 12-core CPU they had already tested extensively in-house.
Funcom’s technical leadership had one clear requirement: consistent, rack-dense, low-power servers, deployed with minimal lead time, hosted in the same physical facility – and i3D.net delivered.
As the game uses the best and newest hardware in the market, the servers had to be ordered 12 weeks in advance, and installed by i3D.net, before Funcom could place the production order. When the first production order was placed, i3D.net provisioned 600+ servers in under 2 hours in US west region. Globally, i3D.net’s average FlexMetal delivery time is just 8.09 minutes, with a 99.84% average API uptime to match.
Funcom’s technical leadership had one clear requirement: consistent, rack-dense, low-power servers, deployed with minimal lead time, hosted in the same physical facility – and i3D.net delivered.
As the game uses the best and newest hardware in the market, the servers had to be ordered 12 weeks in advance, and installed by i3D.net, before Funcom could place the production order. When the first production order was placed, i3D.net provisioned 600+ servers in under 2 hours in US west region. Globally, i3D.net’s average FlexMetal delivery time is just 8.09 minutes, with a 99.84% average API uptime to match.
Today, Dune: Awakening runs primarily on committed capacity, avoiding on-demand scaling for now. But as the game matures, they’re exploring the flexibility that comes with short-term burst workloads.
Funcom’s platform team uses Terraform to provision its infrastructure globally. But they didn’t wait for i3D.net’s Terraform provider to go public, they asked for early access and got it.
Beyond standard automation, Funcom asked for L2 announcements and virtual IP failover support, which not normally found in bare metal platforms.
This customizability helped the team run a resilient architecture while keeping their orchestration logic centralized.
As a high-profile MMO, Funcom’s infrastructure was subject to threat modeling, third-party validation, and real pre-launch attack simulations. Funcom’s parent company, Tencent, performed network-level tests across multiple providers and i3D.net came out on top.
Tencent’s network team tested the DDoS protection and told us outright:
Key reasons:
Funcom’s launch was coordinated through a global virtual war room involving more than 50 engineers, live service partners, and external vendors. However, the room stayed quiet.
When hardware hiccups occurred, i3D.net rebooted or replaced machines very quickly, often even before Funcom’s monitoring flagged it.
According to Bart Kemps, CCO at i3D.net, the privately-owned network of i3D.net and global presence in many data centers across the globe make it possible to offer a low-latency network infrastructure with no downtime to fit Discords’ needs (for an overview of our network, download our global PoP map). It is that combination that results in an excellent quality of service for Discords customers. “Because Discord is growing so fast, we have to anticipate quickly and creatively to provide the best growth plan for the short term and the long term. This requires an agile organization like i3D.net to be there along every step of Discords journey.”
In addition, fast-paced growing organizations like Discord require an agile organization to tackle challenges that come with such rapid growth.
*Mean Opinion Score (MOS) is an industry-standard that rates the voice and image quality of WebRTC transmissions.