In order to justify the capital expenditures, on-premises HPC clusters must run at high utilization rates, often 90% or higher, and this leads to long wait times for important jobs, says Newill. Nevertheless, some organizations found it still made more sense for an organization to expand its HPC system on premises and manage the high-performance servers, storage, and networks themselves, because other cloud-based HPC clusters without bare metal run slower than on-premises.
Why the lack of efficiency for the cloud in the past? Newill identifies one culprit as the cloud monitoring agents found on traditional cloud servers that consume valuable processor cycles, making the servers less efficient. Beyond the amount of computing power an agent uses, it’s their unpredictable timing across hundreds or thousands of cores that kills performance.
“With a CFD job running on a 10,000-core cluster, each time step might take a few hundred milliseconds to calculate,” says Newill. “A monitoring agent, though, is running unsynchronized on all the microprocessors, and CFD has to wait for them all to sync before it can move onto the next time step. If even one agent is running a little behind, the latency for that time step might increase 15% or 50%. That’s a huge, and unpredictable, performance hit.
Newill points out that organizations don’t have to compromise when moving HPC workloads to the cloud: “You can use all the cores on a chip, you can get the same performance as on-premises, but with all of the benefits of the cloud. You can scale up. You can scale down. And it’s all at your fingertips.”
Alan Zeichick is director of strategic communications for Oracle, where he provides insights and analysis on cloud computing and other advanced technologies. Follow him ...
United States Latest News, United States Headlines
Similar News:You can also read news stories similar to this one that we have collected from other news sources.
Source: runnersworld - 🏆 19. / 71 Read more »
Source: BusinessInsider - 🏆 729. / 51 Read more »
Source: latimes - 🏆 11. / 82 Read more »
Source: ELLE Magazine (US) - 🏆 472. / 51 Read more »
Source: Reuters - 🏆 2. / 97 Read more »