The Bandwidth Math Behind Choosing Snowball
~8 min read
A concrete way to reason about when physical transfer beats the network, rather than relying on intuition alone.
The decision between Snowball and network transfer comes down to a straightforward calculation: dataset size divided by realistically sustained available bandwidth gives an estimated transfer time, which can then be compared against how long ordering, loading, shipping, and importing a Snowball device would take (typically one to a few weeks depending on data volume and device count). A useful mental benchmark: even a solid dedicated 1 Gbps connection, sustained perfectly with zero contention (an optimistic assumption in practice), transfers roughly 10TB per day at best — a 500TB dataset would take on the order of 50 days of continuous, uncontended transfer at that rate, versus a Snowball job that could realistically complete in one to two weeks including shipping. This is why exam scenarios and real infrastructure decisions alike treat 'very large dataset, limited timeframe, limited bandwidth' as close to a direct signal toward Snowball, without needing to guess.
💬 Deep Dive with AI
Key points
- •Compare: dataset size ÷ sustained bandwidth (network transfer time) vs. device lead time + shipping (Snowball time)
- •A rough benchmark: even a strong 1 Gbps link moves roughly 10TB/day under ideal, uncontended conditions
- •Snowball jobs typically complete in one to a few weeks including shipping, largely independent of data volume within a device's capacity
- •Large dataset + limited timeframe + limited/shared bandwidth is a strong, calculable signal toward Snowball