~/talk

vSMT-IO: Improving I/O Performance and Efficiency on SMT Processors in Virtualized

on YouTube 20m
Talkat USENIX ATC '20 · 2020-07-15..07-17· [databases]

speakers

name:Weiwei Jia
links:none on file
name:Jianchen Shan
links:none on file
name:Tsz On Li
links:none on file
name:Xiaowei Shang
links:none on file
name:Heming Cui
links:none on file
name:Xiaoning Ding
links:none on file

resources

video:https://youtu.be/pxM5s5QNxKQ
slides:none on file

summary

Proposes context retention on SMT cores: a hardware thread holds a vCPU context while it waits for I/O so the system avoids busy polling and vCPU switches. Implements long-term timeouts and retention-aware symbiotic scheduling to prevent SMT resource starvation, adjusts timeouts dynamically, and raises weights for short retainers. Builds a KVM prototype and measures DBMS, web, AI and Hadoop workloads, reporting up to 88.3% I/O throughput and 123.1% CPU throughput gains.

AI-written summary · about

chapters (10)

  1. 0:00Start
  2. 2:05I/O workloads are mixed with computation workloads in clouds
  3. 7:41Algorithm 1 Context Retention Timeout Adjustment
  4. 12:07Experimental Setup
  5. 12:33Evaluation applications and workloads Application Workload
  6. 12:45Evaluation objectives
  7. 13:06Throughputs of eight benchmark pairs
  8. 14:24Analyzing performance Improvement with DBT1 and MultipleClassify
  9. 15:45Response times of seven benchmarks (two VCPUs time-share a hardware thread)
  10. 17:56Conclusions

related talks

~/search
type to search 6,249 conferences, 138,987 speakers and 185,172 talks
~/help
cmdctrl k
search every conference, speaker and series
/
focus the filter search
h
home
c
conferences
s
speakers
e
series
t
light / dark — the status bar button does the same
?
this help
esc
close this, then clear every filter — or just the search, while you are typing in it

Ignored while you are typing. Pages without a filter rail have no search to focus.