As the picture illustrates, the test system was built around out UCS B-Series servers with EMC CLARiiON storage system with a mix of Fibre Channel drives and state- of-the-art Enterprise Flash Drives (EFDs) to further speed performance. Its important to note that, because the Cisco UCS is architecturally consistent across form factors, the same system could be built with the recently announced C-Series rack servers.
We have published a white paper on the tested deployment that digs into a fair amount of detail on the setup and configuration of the system--highly encouraged reading for anyone that supports Oracle in an enterprise environment. Please note, you will need to register to access the doc (any follow-up from us is on an opt-in basis).
One of the more interesting things things that came out of the testing was the performance data. We ran the cluster through 24h hrs stress tests for both OLTP (order entry) and DSS (sales history). The results included:
Very consistent CPU utilization: around 40 percent on all eight nodes
No saturation levels of any subsystems (CPU, disk, I/O, or networking)
Sustained FCoE-based I/O ranging between 1.8 and 2.0 GB per second, which could be further divided into 1.4 GB per second of Fibre Channel I/O and approximately 450 MB per second of interconnect communication
No occurrence of I/O bottlenecks or wait times
Excellent I/O service times for storage
We attribute much of this consistency to the UCS architecture and its intrinsic 10GB fabric as well as the use of the EMC CLARiiON storage with EFDs. Note, this was not performance testing, there was no config optimization or the like done for this test, we were more interested in understanding how the system handled long-term sustained loads. We are working on performance benchmarking, so stay tuned for those if you are looking for numbers to compare to other solutions out there.
One of the hidden gems at Cisco is the Internet Protocol Journal. The IPJ describes itself as “… intended to serve as an informational and educational resource for engineering professionals involved in the design, development, and operation of public and private internets and intranets. It does not promote any specific products or services, but focuses on issues facing the network designer or operator. The journal carries tutorial articles (“What is…?”) as well as implementation/operation articles (“How to…”). It provides readers with technology and standardization updates for all levels of the protocol stack and serves as a forum for discussion of all aspects of internetworking.”
The quality of the content is quite good and a subscription is free. One of the reasons I bring this up for our readers is the the current edition has a the first part of a two part primer by T. Sridhar on cloud computing. For those of you looking to come up to speed, it is a good vendor-agnostic intro to the topic.
There still continues to be an immense amount of confusion around what exactly constitutes cloud computing. As an example, over the last few days, there has been a spirited debate whether the Sidekick service Microsoft/Danger offered was a cloud-based service or not (I am not going to dig into that here, but I do happen to agree with Chris Hoff’s viewpoint).
A common question I get is if virtualization is an inherent and mandatory component of any cloud solution. I bounced this question up to our CTO’s office to get their take on things. Their perspective:
Cloud computing delivers IT resources on-demand and elastically, and many organizations would like to leverage these capabilities today. Compute virtualization established itself as a way to improve resource utilization, but has other characteristics that make it more broadly relevant to cloud.
To expound on this a bit further, Glenn Dasmalchi, technical chief of staff in the office of the CTO at Cisco, provides a summary of how cloud computing and virtualization are related, and what advantages are afforded to customers. He also touches on the network play with the cloud-virtualization linkage.
With Oracle Openworld running this week it’s a good time to think about the performance of applications over the wide area network (WAN). With the changing business environment IT organizations are re-architecting how they deploy enterprise applications. Oracle applications automate business processes and increase productivity, but in this new network-centric world their performance is being impacted by longer WAN links and inefficient Internet protocols resulting in reduced performance and decreased user productivity.
Globalization and outsourcing has created a workforce that is spread across the world. At the same time, to reduce IT costs resulting from the spread of regional systems, and to take advantage of large scale virtualized systems, organizations are centralizing their application and IT infrastructure in the data center. The result of this trend is that the distance from the end user to the application server has increased dramatically and the connection is often over links with limited bandwidth.