Amazon Web Services has made its home-brewed Arm-powered Graviton2 processors the default for its ElastiCache service.
ElastiCache is AWS’s in-memory-data-store-as-a-service and lets users create Redis or memcached implementations in the Amazonian cloud.
AWS got all excited about ElastiCache on Graviton2 back in December 2019, when it said it had benchmarked Memcached-on-Arm and found 43 percent better performance and lower latency. The cloud colossus therefore said it would run ElastiCache on its own silicon.
The internet giant was true to its word, last week announcing that ElastiCache now supports Graviton2-powered M6g and R6g instance types and that they will deliver “up to a 45 per cent price/performance improvement over previous generation instances.”
As spotted by Australian developer (and former GNOME foundation director) Jeff Waugh, the announcement includes the following sentence:
Graviton2 instances are now the default choice for ElastiCache customers.
ElastiCache-on-Graviton2 is not available in all AWS bit barns, so while the web goliath has declared it is the default CPU microarchitecture, customers still have many x86 options. Bear in mind AWS promotes Graviton2-powered ElastiCache as an upgrade on the basis that it’s faster and offers more network bandwidth than comparable x86-packing instance types.
“ElastiCache Graviton2 instances also come with the latest Amazon Linux 2 which replaces the soon-to-be deprecated Amazon Linux 1,” customers are told. “Furthermore, Graviton2 instances support the latest versions of Redis and Memcached with a seamless upgrade from previous generation instances.”
ElastiCache is an example of AWS behaving more like a SaaS vendor than a cloud. And the processor model a SaaS supplier uses is arguably irrelevant if customers get the software, performance, and price they want. The likes of Salesforce hardly ever reveal what hardware they run on.
Oracle adds Arm-powered servers with up to 160 cores to its cloud – must be why it sunk millions into AmpereREAD MORE
Now AWS is telling its customers it can deliver a superior service with its own silicon.
Which raises the distinct possibility that AWS will do this again for other workloads. Intel and AMD won’t like that, because not many users are ready to run code cut for Arm so the architecture is not yet a huge threat to sales of Xeon or Epyc microprocessors. But plenty of users will be happy to do whatever is needed to cut SaaS costs, because in many cases they won’t have to make major changes.
Game on. ®