900 Petabytes and Not a Watt More
Designing low-power, high-capacity storage in the age of petabyte-class SSDs
SSDs are now available in a wide range of performance-capacity ratios, allowing architects to specify the capacity and performance of their systems independently. Because a storage system’s power follows its bandwidth rather than its capacity, that choice now decides how much of a data center’s power goes to storage instead of GPUs. The largest SSDs hold the most data per watt. But each one takes more data with it when it fails and takes longer to rebuild, so these drives need data protection that scales with their size.
Join VAST Data’s Glenn Lockwood as we use a real 900 PB system, in production today with 8,008 SSDs of 122 TB each, to explore the power-capacity design space opened up by high-capacity SSDs. Starting from its actual configuration, we will discuss what storage bandwidth costs in GPU power and how much power next-generation SSDs could save. We will then run ten years of simulated drive failures against this exact system to show which data protection schemes stay safe as drive capacities approach a petabyte. Finally, we will share what systems of this scale have taught us about detecting failures and coordinating recovery across hundreds of servers.
What you’ll learn:
- Where do 122 TB, 245 TB, and larger SSDs fit into the design space of hyperscale data platforms?
- What special considerations are required to protect against data loss when designing high-capacity, low-power systems with these SSDs?
- When does adding more redundancy stop making data safer, and what matters more after that?
We appreciate your interest! Join us on Wednesda, December 9th at 12 pm ET / 9 am PT.