You are using iSCSI to transfer files between a 10 Gigabit Ethernet storage system and a 1 Gigabit Ethernet server. The performance is only approximately 700 Mb/s and output drops are occurring on the

WRED (Weighted Random Early Detection) monitors average queue depth and randomly discards packets as congestion builds, triggering TCP's congestion control mechanism at the sending storage system to reduce its window size and transmission rate. This prevents tail-drop scenarios where the entire queue fills and all packets are lost, resulting in more stable throughput than reactive drop policies. It is cost-effective because it requires only a QoS configuration change on the switch rather than a hardware upgrade.

A 1 GB output queue would delay drops but not prevent them - it causes severe bufferbloat by increasing latency without signaling TCP to slow down, and 1 GB of switch buffer memory is neither practical nor cost-effective.

The TCP Nagle algorithm coalesces small segments to reduce overhead, but iSCSI workloads require Nagle to be disabled for optimal block storage performance; enabling it on the receiver would not resolve queue overflow from a speed mismatch.

CIFS (Common Internet File System) is a file-level sharing protocol and is not an equivalent replacement for iSCSI block storage; switching protocols does not address the underlying 10G-to-1G bandwidth mismatch causing the drops.