~/talk

More Than Capacity: Performance-oriented Evolution of Pangu in Alibaba

on YouTube 15m
Talkat FAST '23: 21st USENIX Conference on File and Storage Technologies · 2023-02-21..02-23· [networking]

speakers

name:Qiang Li
links:none on file
name:Qiao Xiang
links:none on file
name:Yuxin Wang
links:none on file
name:Haohao Song
links:none on file
name:Ridi Wen
links:none on file
name:Wenhui Yao
links:none on file
name:Yuanyuan Dong
links:none on file
name:Shuqi Zhao
links:none on file
name:Shuo Huang
links:none on file
name:Zhaosheng Zhu
links:none on file
name:Huayong Wang
links:none on file
name:Shanyang Liu
links:none on file
name:Lulu Chen
links:none on file
name:Zhiwu Wu
links:none on file
name:Haonan Qiu
links:none on file
name:Derui Liu
links:none on file
name:Gexiao Tian
links:none on file
name:Chao Han
linkedin:chaohandata
name:Shaozong Liu
links:none on file
name:Yaohui Wu
links:none on file
name:Zicheng Luo
links:none on file
name:Yuchao Shao
links:none on file
name:Junping Wu
links:none on file
name:Zheng Cao
links:none on file
name:Zhongjie Wu
links:none on file
name:Jiaji Zhu
links:none on file
name:Jinbo Wu
links:none on file
name:Jiwu Shu
links:none on file
name:Jiesheng Wu
links:none on file

resources

video:https://youtu.be/Twss7x3Ib2Q
slides:none on file

summary

A global unified storage system evolves to deliver 100‑microsecond average latency by deploying a user‑space storage OS (USS OS), RDMA from 25Gbps to 100Gbps, and a redesign of file layout and RPCs to cut RPCs and context switches. Remote direct cache access (RDCA) bypasses DDR, CPU offload uses FPGA compression and NIC CRC, background TCP traffic shifts to RDMA to cut memory use ~75%, and network traffic amplification 66.3→2.9 yields a measured 6.1× throughput uplift and p99 write latency ≈1 ms.

AI-written summary · about

chapters (10)

  1. 0:00Intro
  2. 0:26Pangu: The Unified Storage Platform in Alibaba Cloud
  3. 2:35Storage Architecture based on Pangu2.0
  4. 3:30Performance Design Challenges for Pangu2.0
  5. 5:47Full Stack Optimization for Latency
  6. 7:28Methods to Guarantee Performance SLA
  7. 9:20Latency and SLA Evaluation
  8. 10:06Higher I/O Efficiency, Better Performance
  9. 11:02Breaking Network Bottleneck to Improve I/O Efficiency
  10. 11:54Breaking Memory Bottleneck to Improve I/O Efficiency

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.