You are using iSCSI to transfer files between a 10 Gigabit Ethernet storage system and...

The performance degradation is caused by output queue drops on the 1 Gbps server port. Increasing the queue size allows more data to be buffered during transient congestion periods, preventing packet loss and retransmissions which significantly degrade iSCSI performance.

iSCSI is sensitive to packet loss due to its reliance on TCP.

Increasing the queue depth is a simple, cost-effective method to absorb bursts.

Changing to CIFS (A) is irrelevant since CIFS would suffer similar TCP congestion.

WRED (C) intentionally drops packets during congestion, which is undesirable here.

Enabling TCP Nagle (D) would likely introduce latency and reduce throughput for large data transfers.

—