Shrink Your Blast Radius, Not Your Ambitions

The blast radius can be an issue in distributed storage solutions where a single point of failure, such as a shared solid-state drive (SSD) or server, can impact multiple servers at once. This makes the storage unavailable to all connected servers. While disaggregated storage via technologies like NVMe-oF™ is a common strategy to improve efficiency, it can also impact the blast radius. For example, if multiple drives are shared from a single server over NVMe-oF, a failure in that server can make the entire rack of application servers unusable.

What is the Blast Radius?

The blast radius refers to the scope of disruption caused by a single hardware failure—like an SSD or server—within a storage architecture. In traditional direct-attached storage (DAS) configurations, a single drive failure affects only the server it’s attached to. But as organizations move toward disaggregated architectures using NVMe-oF, the blast radius can become exponentially larger.

Imagine this scenario: You have racks filled with servers, each accessing shared NVMe SSDs over a high-speed fabric. These SSDs are partitioned and shared across multiple application hosts. If a single SSD fails, every application server relying on it could be affected—impacting an entire rack or cluster. That’s a big blast radius—with potential to disrupt SLAs, delay rebuilds, and reduce performance across your entire application stack.

The Challenge with Legacy DAS Storage

Organizations often use local NVMe SSDs in servers for low-latency and high-speed application performance. However, this setup has its drawbacks. Since the storage is directly attached, it’s not protected by default, requiring the applications themselves to handle data replication or protection. This can be a problem if the application wasn’t designed for this task. Even with modern, self-replicating applications like MongoDB or Cassandra, a drive or server failure necessitates a full rebuild over the network, which can take many hours for common 8-16 TB drives. This can lead to degraded performance and potential service level agreement (SLA) violations.

Disaggregation with NVMe-oF Is Worth It (Despite the Risk)

NVMe SSDs are now available in 16TB capacities and beyond, with 30TB and 64TB models increasingly common. These drives offer far more performance (IOPS and throughput) than any single server can consume. Sharing them via disaggregated storage architectures—using NVMe-oF—enables better resource utilization, improved scalability, and lower TCO.

Benefits include:

Higher efficiency: Share underutilized capacity and IOPS across servers.
Lower cost: Reduce hardware footprint by pooling storage resources.
Faster scaling: Add compute and storage independently, based on need.

However, with traditional open-source NVMe-oF implementations or simple RAID schemes, failure of a shared storage node or SSD can take down multiple application servers, which doesn’t solve the blast radius problem.

How Lightbits Solves the Blast Radius Problem

Lightbits Labs has reimagined NVMe-oF storage to address this exact issue. Our software-defined storage is designed to eliminate the risks associated with drive or server failure while still delivering the full benefits of disaggregated NVMe storage.

Here’s how:

Granular, Intelligent Rebuilds: In case of an SSD failure, Lightbits intelligently rebuilds only the data that has been allocated, rather than the full capacity of the drive. This significantly shortens recovery times—transforming hours-long rebuilds into minutes or seconds—and limits impact to only the relevant workloads.
Rebuilds Happen Within the Target System: Unlike many open-source NVMe-oF solutions that require data to be rebuilt across the network, Lightbits performs rebuilds inside the target server. This reduces bandwidth overhead, minimizes recovery latency, and avoids impacting network performance for active applications.
Multipath Redundancy with ANA Support: Lightbits includes standards-based Asymmetric Namespace Access (ANA), enabling multipathing across storage targets. If a server or controller goes down, application servers automatically redirect traffic to another available target. The failover is seamless and ensures zero data loss and minimal disruption.
Redundant, Clustered Architecture: Lightbits supports high-availability clusters, ensuring no single point of failure in the data path. Combined with erasure coding, advanced data placement, and policy-based QoS, the system maintains service continuity even in the face of hardware failure.

Lightbits offers NVMe-oF that is easy to deploy and highly performant. We invented NVMe/TCP; it’s natively designed into the software, avoiding the complexity of RDMA while delivering line-rate performance and sub-millisecond latencies. It integrates seamlessly with modern orchestration platforms like Kubernetes and OpenStack, and supports cloud-native environments and VM-based workloads alike.