During capacity planning for a new streaming pipeline, an architect estimates that connected IoT devices will generate 250 MB of data every second. The data is written from edge gateways to a six-node on-prem object-storage cluster. Each storage node currently uses a single 1 GbE network interface, has 8 vCPU cores, 16 GB of RAM, and SATA HDDs that sustain about 200 MB/s sequential writes. During a pilot test, the ingestion queue grows quickly; monitoring shows disk-write utilization at 65 percent, CPU under 40 percent, RAM mostly free, but every network port pinned above 95 percent. Which single change to the resource sizing is MOST likely to remove the immediate bottleneck while keeping additional cost low?
Replace each 1 GbE adapter with a 10 GbE network interface.
The key to solving this bottleneck is identifying the most constrained resource. The scenario indicates that each node's 1 GbE network interface is operating at over 95% capacity. A 1 GbE link has a theoretical maximum throughput of 125 MB/s (1 gigabit / 8 bits per byte). Operating at 95% means each link is handling about 119 MB/s. While the cluster's total ingest rate of 250 MB/s averages to only ~42 MB/s per node, object storage systems generate significant internal network traffic by replicating data chunks across nodes for redundancy. This replication traffic, when added to the initial ingest traffic, is saturating each node's network connection. In contrast, CPU utilization is low (<40%), RAM is mostly free, and disk utilization is manageable (~65%), so none of these are the immediate bottleneck. Upgrading from a 1 GbE to a 10 GbE adapter increases each node's network capacity to approximately 1,250 MB/s, providing ample bandwidth to handle both ingest and replication traffic. This is also typically the most cost-effective upgrade compared to adding more CPU or RAM (which are not constrained) or switching to much more expensive NVMe storage (which is not the current bottleneck).
Ask Bash
Bash is our AI bot, trained to help you pass your exam. AI Generated Content may display inaccurate information, always double-check anything important.
Why does upgrading from a 1 GbE to a 10 GbE network interface solve the bottleneck?
Open an interactive chat with Bash
What role does data replication play in increasing network utilization?
Open an interactive chat with Bash
Why are CPU, RAM, and disk resources not the bottleneck in this scenario?