mirror of
https://github.com/deepseek-ai/3FS.git
synced 2026-11-05 00:47:47 +00:00
[GH-ISSUE #98] Unable to Deploy 3FS Without RDMA – Consider Virtualizing an RDMA Network #55
Labels
No labels
pull-request
No milestone
No project
No assignees
1 participant
Notifications
Due date
No due date set.
Dependencies
No dependencies set.
Reference
deepseek-ai/3FS#55
Loading…
Add table
Add a link
Reference in a new issue
No description provided.
Delete branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Originally created by @nfyang-nexus on GitHub (Mar 5, 2025).
Original GitHub issue: https://github.com/deepseek-ai/3FS/issues/98
In our standard network deployment environment, it is not possible to deploy 3FS directly because RDMA networking is mandatory. Therefore, we can consider virtualizing an RDMA network environment to meet the requirements.
Steps to Set Up a Virtual RDMA Environment
1. Install Required Packages
2. Load the rxe Kernel Module
3. Verify That the Module Is Loaded
Expected output:
4. Check Available Network Interfaces
Example output:
• lo: Loopback interface for local communication (127.0.0.1), not a physical NIC.
• enp1s0: The primary network interface (Ethernet), status UP, MAC address 52:54:00:71:4b:d8.
• docker0: Docker virtual bridge, currently DOWN, meaning no active container connections.
5. Add enp1s0 as an RDMA Device
6. Verify RDMA Device Creation
If successful, you should see the new rxe0 device, which serves as the virtual RDMA version of enp1s0.
Run 3FS with the Virtualized RDMA Network
Once the virtual RDMA environment is set up, you can proceed with starting 3FS:
This should now work as expected in an environment where physical RDMA hardware is unavailable.
@SSmallMonster commented on GitHub (Mar 5, 2025):
so great, this makes me 'can touch 3fs'!
@SSmallMonster commented on GitHub (Mar 5, 2025):
If anyone occurs err like:
modprobe: ERROR: could not insert ‘rdma_rxe’: Invalid argument, try to uninstall mlnx_ofed package and reboot(to reload modules).Some symbols build version in
rdma_rxeis conflict with the version in the kernel moduleib_corewhich imported from mlnx package.@zartbot commented on GitHub (Mar 5, 2025):
AlibabaCloud provide all region and all az support for RDMA on 8th generation generic computing instance with high performance. Here is a chinese deployment guide https://mp.weixin.qq.com/s/Ri9-sjsaoQ3LUE2-dTAN_g , just FYI.
@xiaoyaoyouyue commented on GitHub (Mar 6, 2025):
I install the rdma-core, and reboot, but modeprobe FATAL , do you know the reason ? thanks!
root@ubuntu:
# modprobe rdma_rxe# sudo apt install -y rdma-core infiniband-diags ibverbs-utilsmodprobe: FATAL: Module rdma_rxe not found in directory /lib/modules/5.15.0-1073-kvm
root@ubuntu:
Reading package lists... Done
Building dependency tree... Done
Reading state information... Done
rdma-core is already the newest version (39.0-1).
ibverbs-utils is already the newest version (39.0-1).
infiniband-diags is already the newest version (39.0-1).
0 upgraded, 0 newly installed, 0 to remove and 41 not upgraded.
root@ubuntu:~# modprobe rdma_rxe
modprobe: FATAL: Module rdma_rxe not found in directory /lib/modules/5.15.0-1073-kvm
@SSmallMonster commented on GitHub (Mar 6, 2025):
try to install by
apt install linux-modules-extra-$(uname -r), the module is provided by the linux-modules-extra packages.@freedomsunny commented on GitHub (Mar 7, 2025):
great!
@xiaoyaoyouyue commented on GitHub (Mar 8, 2025):
(base) root@ubuntu:
/3fs# apt install linux-modules-extra-$(uname -r)/3fs#Reading package lists... Done
Building dependency tree... Done
Reading state information... Done
E: Unable to locate package linux-modules-extra-5.15.0-1073-kvm
E: Couldn't find any package by glob 'linux-modules-extra-5.15.0-1073-kvm'
(base) root@ubuntu:
命令尝试如上,另外我尝试了使用升级内核的方式,结果系统升级挂了
@xiaoyaoyouyue commented on GitHub (Mar 11, 2025):
使用generic的系统可以了
@Icedroid commented on GitHub (Mar 13, 2025):
I run in docker , but mgmt not start successfully. How cloud I do to resolve it?
@ghost commented on GitHub (Mar 23, 2025):
ibdev2netdev This file is required
@SimonCqk commented on GitHub (Apr 25, 2025):
Is there anyone works on native TCP network environment?