You can see all the attendees[1] and agenda[2]. People come from various companies, familar and new faces, experting at different project. It's intersting to see the "maillist" guy in person.html
[1] https://www.china-lsf.org/clsf_2017_invitation_list.html
[2] https://www.china-lsf.org/clsf_2017_agenda.htmlsession
Lustre is a very heavy user of ext4, contributing many important features for ext4, like malloc, mmp, large EA,project quota. Shilong Wang talked about the background of these features and performance issues, what they did and their achievments.ide
It seems an exploring project in Huawei now to unify the use of memory from CPU side and GPU side.oop
This is a very impresive presention to me. Junhui, half year ago, I saw he contributed many patches for multipath, now he said he is doing ceph now. In this session, he presented how they use bcache on ceph to improve the performance. I tink his PPT is a very helpful example to handle IO performance issue.ui
This is a more research topic about using Formal verification to verify FS.this
I've seen this topic from him from other different meetings, ceph Day, CLK. SPDK, DPDK, SDD disk, ceph, RDMA, all these thingys, usually being brought up together, the goal is to achive very high IOPS and low latency.idea
I thought this project is just about reusing the kernel code in userspace program, like the "list_head" implemetation. However, as Gang said in his summary, its intend is to run kernel in userspace, well, I don't understand the reason.spa
The name BloodTest tell the idea behind this project. Like in the hospital, doctor only need a little of the patient's blood to analyze the healthy situation. He used this tool to help troubleshooting the transient performance decline. Apparently, it's a common problem which brought up much more discussion beyond the tool itself.rest
Yu Chao is a very principle guy in f2fs community, who formally worked at Samsung. The interesting part is 5 out 6 core developers are from Huawei now, except one mainainter is working at Google now. He shared his PPT you can see detailed updates there.code
Lei ming is a very active developer in block layer, who formally worked at Cannonical. Bvec is a vector of (page, offset, size), which is used in bio to gather/scatter IO, but as you see, one bvec cannot stride across more than one page. However, large IO become normal nowadays. Multipage bve will decrease the amount of bve, which will reduce the overhead to process a bio.
THP (Transparent Huge Page) on swap and fs seems a huge ongoing work in upstream. Huang ying is very dedicated at swap optimization that the purpose is to help sale more "virtual" memory on the cloud using fast SSD as swap. The huge page topic is very interesting, you can refer to his PPT for the upstream status.
I don't know much about container, but the most intersting part to me is, as said by the presenter, intel will cooperate with a startup company Hyper[1]