Hi, SPDK & Netdev,
When running fio with SPDK iscsi target,
we find the localhost interface has a bad performance,
contrast to access from the remote node.
After upgrade the kernel to 5.1, this issue still exists.
There are some fio data of 4k randread, QD=128, jobs = 1:
iSCSI server run on the NODE1, fio run on NODE1 or NODE2.
Fio + localhost(NODE1) + openiscsi + SPDK + NVMe
Fio + localhost(NODE1) + openiscsi + SPDK + NVMe (5.1 kernel)
Fio + remote node(NODE2) + openiscsi + SPDK + NVMe
Connect localhost target with another interface ip (e.g. 10.0.xx.yy),
the performance is poor. The kernel driver could recognize this is a
local access, and use loopback directly.
Setup a new namespace, and add another interface in this ns, and iSCSI
target listen on this interface.
Openiscsi connects from the current host, the performance is good.
1. Is this a known issue about the loopback? Any workaround?
2. If this is no solution about loopback, how do we force the local
interface(127.0.0.1, and other nic interfaces) over the wire, not
directly pass in the host?
The SmartX email address is only for business purpose. Any sent message
that is not related to the business is not authorized or permitted by