Alleviating flow interference in data center networks through fine-grained switch queue management

Modern data centers need to satisfy stringent low-latency for real-time interactive applications (e.g. search, web retail). However, short delay-sensitive flows generated from these applications often have to wait a long time for memory and link resource occupied by a few of long bandwidth-greedy flows because they share the same switch output queue (OQ). To address the above flow interference problem, this paper advocates more fine-grained flow separation in the switches than traditional OQ. We propose CQRD, a simple queue management scheme for data center switches, without change to the transport layer and requiring no coordination among switches. Through simulations, we show that CQRD can reduce the flow completion time (FCT) of short flows by more than 25% in a single switch and up to 50% in a multi-stage data center network, only at the cost of a minor goodput decrease of large flows. Additionally, just a 50% deployment of CQRD in top-of-rack (ToR) switches can lead to a ~10-24% FCT reduction of short flows. Moreover, CQRD can improve short flows' FCT by ~30-40% from OQ switches, using DCTCP (Alizadeh et¿al., 2010) 2 transport in DCN. Furthermore, we validate the feasibility of CQRD approach by implementing an 8 × 8 logical CQRD switch through simply changing the configurations of existing commodity switches. Also, we use a small testbed experiment to verify the implementation and the effectiveness of CQRD to alleviate flow interference in real environment.