Cisco Cisco UCS B260 M4 Blade Server White Paper

Página de 11
© 2016 Cisco and/or its affiliates. All rights reserved. This document is Cisco Public information. 
Page 6
Cisco UCS and Dell PowerEdge M1000e: A Comparison
June 2016
use of bandwidth and more headroom in the event that the application requires 
bandwidth from other modalities. Cisco UCS uses configurable quality-of-service 
(QoS) controls so you can assign and reassign bandwidth priorities, and it enables 
your applications to dictate connectivity and bandwidth. 
For example, Figure 2 shows three types of traffic: virtual machine movement 
(green), SAN (yellow), and LAN (blue). Suppose the total available bandwidth is 10 
Gbps, and you set a bandwidth guarantee of 3 Gbps for both the LAN and SAN and 
2 Gbps for virtual machine movement traffic. If each modality offers 3 Gbps of actual 
traffic, the bandwidth allocation would look like time-step one (t1) in Figure 2, giving 
you a spare 1 Gbps. If the LAN has a burst of traffic to 4 Gbps as shown in time t2 
and the bandwidth is available, LAN traffic will be allowed to use that bandwidth. 
Time t3 shows a burst of 6 Gbps of LAN traffic. Because the SAN is using only 2 
Gbps of its allocated bandwidth, the LAN traffic is allowed to use 6 Gbps because it 
can “borrow” unused bandwidth from the SAN. 
Deterministic Latency
Cisco UCS unified fabric reduces latency and makes latency times more consistent. 
Cisco UCS fabric interconnects centrally manage network traffic within Cisco UCS 
as well as the traffic in and out of the system. For traffic that is moving from one 
chassis to another in the same Cisco UCS domain, there is no need to exit the 
system and send packets through another external switch (see path A in Figure 
3). This central connectivity demonstrates one of the ways in which Cisco UCS 
functions as a single virtual chassis and provides the flexibility to place workloads 
anywhere in the system with assurance of consistent network performance. 
This capability is important because it simplifies the placement of virtual machines 
in virtualized and cloud environments: regardless of location, virtual machines 
experience the same latency, removing the network as a placement constraint. 
When trying to reduce latency between servers, the best-case scenario for any 
vendor is a single network hop for data communicated between servers within 
Offered Traffic
t1
t2
t3
t1
t2
t3
Actual 10 GE Bandwidth Allocation
3 Gbps 3 Gbps 2 Gbps
3 Gbps
4 Gbps 6 Gbps
3 Gbps
2 Gbps
Virtual Machine
Movement Traffic
3 Gbps
3 Gbps
2 Gbps
Storage Traffic
3 Gbps
3 Gbps
6 Gbps
LAN Traffic
4 Gbps
3 Gbps 3 Gbps 2 Gbps
Figure 2 Cisco UCS Dynamic Traffic Allocation Example
Cisco’s test results are summarized