TECH Signal 389
Depot reimplemented LVM without crash safety for roughly 100x faster microVM volume allocation
↗
Depot replaced LVM with a custom in-process allocator using device-mapper directly, trading crash safety guarantees for roughly 100x faster volume operations on ephemeral microVM workloads.
LVM's volume group-wide locking makes it unsuitable for workloads requiring high-frequency parallel volume operations. For ephemeral microVM workloads where crash safety provides no value, trading those guarantees for massive speedup is a rational engineering decision. This demonstrates when established infrastructure guarantees become pure overhead.
Written by elseif from the cluster below · every claim links back to a sourceThe three things worth knowing
LVM's VG-wide lock holds for roughly 100ms per operation, blocking the 200 parallel operations per second that Depot's microVM hypervisors require.
Crash safety guarantees from LVM are unnecessary for Depot's use case because hypervisor crashes discard ephemeral workloads and lose volatile memory anyway.
Depot's in-process allocator manages a memory pool and uses device-mapper directly, achieving roughly 100x speedup over LVM for the same operations.
THE CLUSTER