Hybrid Transmission Time Intervals for TCP Slow
Start in Mobile Edge Computing System
Siqi Mu
∗
, Shiyu Wu
∗
, Ming Liu
∗
, Naizheng Zheng
†
, Hong Zhou
†
, Yong Teng
†
,
Xia Chen
‡
, Qi Zhang
∗
, Yawen Zheng
∗
, Zhangdui Zhong
∗
, Hao Guan
†
∗
State Key Lab. of Rail Traffic Control and Safety, Beijing Jiaotong University, Beijing, China
†
Nokia Bell Labs, Beijing, China
‡
School of Electronic and Information Engineering, Beijing Jiaotong University, Beijing, China
Email: {musiqi, 16120142, mingliu, xchen, 17120166, 17120454, zhdzhong}@bjtu.edu.cn
{naizheng.zheng, hong.1.zhou, yong.teng, hao.guan}@nokia-sbell.com
Abstract—In this paper, a novel scheduling scheme with
hybrid transmission time intervals (TTI) is proposed to im-
prove the performance of Transmission Control Protocol (TCP)
transmission in Mobile Edge Computing (MEC) systems. Two
use cases with diverse service requirements, i.e., low latency
communication (LLC) and enhanced mobile broadband (eMBB),
are considered and originally assigned to separate bands with
short TTI of 0.143 ms and long TTI of 1 ms, respectively. With
the proposed scheduling scheme, eMBB traffic at TCP slow start
stage can be scheduled with short TTI and utilize the remaining
resource after scheduling LLC users, thus speeding up the slow
start procedure at the initial transmission phase due to faster
feedback of acknowledgement. After a given time threshold, while
the throughput probably approaching saturation, eMBB traffic
switches to its own band with long TTI to lower transmission
overhead. System level simulations demonstrate that the proposed
solution can achieve high performance improvement for eMBB
users while not affecting LLC users in terms of user throughput
and transmission latency, particularly when the eMBB traffic
load is heavy compared to its available bandwidth. The switching
time threshold should be flexibly adapted to control the amount of
eMBB traffic allocated with frequency resources initially reserved
for LLC users, such that the eMBB traffic can benefit from
utilizing the vacant resource, thereby achieving the potential gain.
I. INTRODUCTION
Research on the 5G New Radio (NR), including radio
network architecture and air interface techniques, are gaining
more attention in both academic and industrial communities,
for supporting a large variety of services and deployment use
cases, such as enhanced mobile broadband (eMBB), massive
machine-type communication (mMTC) and ultra-reliable low
latency communication (LLC) [1]–[3]. Currently, traffics from
remote central servers to mobile user equipments (UEs) proba-
bly need to cross wide area networks (WANs), suffering packet
loss due to network congestions. Moreover, the long-distance
transmission may well result in high latency, and thus fails
to satisfy the quality of service (QoS) requirements of eMBB
and URLLC.
Mobile Edge Computing (MEC) has been a promising
solution to this issue. Hot-hit contents and latency-critical
application can be cached and processed on the powerful edge
servers. And the content files or processing results are fetched
via just one hop over the air interface, thereby achieving high-
bandwidth and low-latency requirements for eMBB and LLC
services. Meanwhile, the Internet transport protocols such as
the commonly used Transmission Control Protocol (TCP) [4]
can be exploited to guarantee reliable, ordered, and error-
checked delivery in this edge computing network architecture.
However, when adopted over radio link, the throughput of TCP
may decrease dramatically as it cannot differentiate between
the losses caused by wireless transmission errors and network
congestion due to high load. This misinterpretation by TCP
leads to an over-reduced of congestion window size and lower
throughput.
On the other hand, the flexible physical layer design in 5G
NR [1], especially the agile frame structure design [5], offers
more freedom for the scheduling functionality to facilitate
user-centric scheduling. 5G NR allows to schedule the users
with variable transmission time intervals (TTIs) as proposed in
[5]–[7]. In this case, radio resources allocation is more flexible,
and hence can be optimized in coherence with users’ variant
radio conditions and diverse QoS requirements. For instance,
scheduling LLC users with short TTIs to achieve low latency,
while accepting the penalty of higher relative control channel
overhead. The recent studies in [8], [9] shows the benefits
of variable TTIs for LLC traffic. Similarly, scheduling with
variable TTI sizes also provides advantages for eMBB traffic
[7], as shorter TTIs can be utilized to accelerate the TCP
closed-loop flow control mechanisms.
In this paper, a hybrid transmission scheme using two TTIs
of different length on separate frequency bands is proposed
to enhance the performance of an MEC system with mixed
traffic of LLC and eMBB services, in which TCP is used
to provisioning of reliability of both services. Specifically, a
shorter TTI is adopted for LLC traffic and eMBB traffic at the
slow start phase to speed up the TCP interaction to and from
UEs and achieve faster recovery to a higher throughput from
slow start. Then a longer TTI is adopted when the throughput
approaches saturation. System level simulations demonstrate
that the proposed solution can improve performance of eMBB
users while not affecting the LLC users in terms of user
throughput and latency when switching time threshold, current
traffic condition and available bandwidth are well matched.
The rest of the paper is organized as follows. Section II
provides the system architecture and problem description. The
proposed scheme is described in Section III. System-level