Why Cisco UCS Rocks #6: Awesome Performance
And of Intel Xeon processor servers, no other server has better base performance:
With versatile Intel Xeon processors, Cisco UCS demonstrates performance breadth by setting records for raw CPU power. . . . Although all vendors have access to the same powerful Intel Xeon processors, only Cisco unleashes their power to accelerate application performance. — Cisco UCS Performance Brief
Cisco says it this way:
Best for performance. Best for scalability. First published vSphere 5.1 results. Cisco’s continued leadership in cloud computing demonstrates the value of good infrastructure for cloud computing environments. — Cisco UCS Performance Brief
There is actually a VMware Horizon benchmark, called VDImark, and the results are, again, expected. Because UCS is designed from the chassis up to support virtual workloads by abstracting the configuration from the underlying hardware, it is incredibly powerful for virtual desktop infrastructures.
Organizations everywhere can gain fast results for virtual desktop deployments
using Cisco Unified Computing System™ (Cisco UCS) and the new Intel Xeon
processor E5-2600 v2 family. Cisco UCS can be deployed quickly in any data
center, delivering virtual desktop infrastructure (VDI) in minutes and industry-leading response times to real-world desktop workloads.
— Cisco VMware View Performance Brief
But in actuality Cisco UCS still outperforms Dell, HP, and IBM when it comes to East-West (server-to-server) network latency.
One of the ways to accomplish better throughput for networking comes as part of the abstraction layer that is built into UCS. Because my network interface cards are virtual, I can provision them in any number of ways: to the blade directly, to the blade for the hypervisor, or passed through the blade directly the VM, what is known as VM-FEX.
Gain 37 percent greater network throughput using the Cisco® Data Center Virtual Machine Fabric Extender (VM-FEX) compared to the VMware Virtual Switch (vSwitch).
Decrease round-trip times by 40 percent to accelerate performance of transactional applications.
Using the same amount of bandwidth, data sent using the Cisco virtual interface card (VIC) with Cisco Data Center VM-FEX uses 41 percent fewer CPU resources than the VMware vSwitch, freeing CPU cycles to deliver better application performance.
A system’s peak performance is essentially limited by its slowest component. Intel® Xeon® processor performance is becoming ever faster, and the capacity and performance of the rest of the system must keep pace to avoid memory and I/O bottlenecks that limit performance. The Cisco UCS VIC 1240 is the exception, delivering excellent performance for both IP networking and storage traffic. The results demonstrate how Cisco supports a better balance of resources today and has the capacity to provide even greater bandwidth as bus speeds increase.
— Cisco Red Hat Enterprise Linux Performance Brief
Peak throughput was 38.7Gbps at 60% of line rate speed with multiple TCP traffic flows.
VIC 1240 Expander (from 4 channels to 8 channels) increased throughput 94% (almost linear scaling)
The VIC 1240 configured for HBA delivery yielded 7.5GBps (that's Bytes) of read/write peak throughput.