Tencent Cloud and Intel Showcase Cube Sandbox Agent Execution
Tencent Cloud says Cube Sandbox starts in about 48 milliseconds, gives each sandbox an independent Guest Kernel, and adds less than 5 MB of overhead per instance. [[fact:fact-shkvt1pdc]] In the tested configuration, software LZ4 and IAA Deflate both achieved about 2.1:1 compression, while IAA compression and decompression latency measured 0.51 milliseconds and 0.08 milliseconds, respectively. [[fact:fact-yyqu6wlup]] Cube Sandbox uses eBPF for network isolation, denies access by default and allows it through a whitelist, while a Layer 7 proxy injects credentials into outbound requests. [[fact:fact-9sw7c49a1]] Joint testing showed that as the core count rose from 48 to 144, concurrent creations increased from 100 to 250, with average creation latency at about 200 milliseconds. [[fact:fact-yx98v6wnd]] A separate joint test said Cube Sandbox on Xeon 6+ processors supported 350 concurrent sandbox-creation requests at 200-millisecond average creation latency, with concurrency density 1.46 times that of other mainstream test platforms. [[fact:fact-8p7rqtce1]] Hu Yong, software technology director at Intel's Data Center Group, estimated a memory shortfall of about 2 GB per vCPU in an example configuration. [[fact:fact-b4zlmznvq]]