diff options
Diffstat (limited to 'results/classifier/zero-shot/108/performance/2848')
| -rw-r--r-- | results/classifier/zero-shot/108/performance/2848 | 28 |
1 files changed, 28 insertions, 0 deletions
diff --git a/results/classifier/zero-shot/108/performance/2848 b/results/classifier/zero-shot/108/performance/2848 new file mode 100644 index 000000000..27c89cc86 --- /dev/null +++ b/results/classifier/zero-shot/108/performance/2848 @@ -0,0 +1,28 @@ +performance: 0.965 +graphic: 0.758 +device: 0.691 +semantic: 0.657 +vnc: 0.558 +socket: 0.461 +other: 0.405 +PID: 0.317 +network: 0.305 +boot: 0.303 +KVM: 0.214 +files: 0.168 +permissions: 0.166 +debug: 0.132 + +i386 max_cpus off by one +Description of problem: +X86 VMs are currently limited to 255 vCPUs (`mc->max_cpus = 255;` in `pc.c`). +The first occurrence i can find of this limit is in d3e9db933f416c9f1c04df4834d36e2315952e42 from 2005 where both `MAX_APICS` and `MAX_CPUS` was set to 255. This is becoming relevant for some people as servers with 256 cores become more available. + +**Can we increase the limit to 256 vCPUs?** +I think so. + +Today, the APIC id limit (see `apic_id_limit` in `x86-common.c`) is based on the CPU id limit. +According to the a comment for `typdef uint32_t apic_id_t;` (see `topology.h`), we can have 256 APICs, but more APICs require x2APIC support. +APIC seems to be no hindrance to increase max_cpus to 256. + +**Can we increase the limit to 512?** Maybe not? We need x2APIC support of which i have no clue. Also there is always a performance risk of exceeding the size at which current data structures work efficiently. |