In the following, with two simple benchmarks, we would like to demonstrate the advantages of using a local Zettar ZCloud Sandbox as an integral part of your cloud software development and QA process, as opposite to "go live" right away as have been done by most up to now.
We employed a data set that consists of 667 files. Their size vary from 0 byte to 1.04MB. The types of these files also vary: software source codes, Microsoft Office document, different kinds of multimedia files.
In conducting the benchmark, We used the following hardware and software combinations. They were chosen to be both realistic and representative.
a Xen domU, 512MB RAM, 16GB disk space, a single 1.9GHz VCPU, Ubuntu 9.04 32bit server OS
a Xen domU, 512MB RAM, 16GB disk space, a single 1.9GHz VCPU, Ubuntu 9.04 32bit server OS.
The two aforementioned Xen domUs are hosted on the same physical server, a SUN x4140, that is colocated in a hosting data center. The physical server has an AMD quad-core 1.9Ghz Opteron CPU, 2 10000 RPM 68GB SAS HDs in RAID 1, and 2 10000 RPM 136GB HDs in RAID 1, connected to a 100Mbps switched Ethern port. The switch itself is connected to the Internet full time via a 10Mbps port. It runs 8 identically configured Xen domUs. We also used an Amazon EC2 small instance as a client. Its spec: EBS, 1.7GB RAM, 150 GB data storage, 10GB root partition, Ubuntu 9.04 32bit OS. The main reason is to use a small instance is to match our more modest hardware and Xen domUs as closely as possible.
We wanted to use AWS S3 client software that had been widely used and highly rated, and could be easily scripted. Thus, the following two were selected:
This tool is published by Amazon Web Services, so we regard it as the baseline tool for all our benchmarking activities.
Quite efficient. Note that a few settings of this AWS S3 client software can have significant impact to a benchmark's outcome.
The above also demonstrate the fact that an efficient server must be paired with an efficient and well tuned client for the most optimal performance.
We also benchmarked the ZCloud Sandbox software using two KVM guests set up on a notebook running Ubuntu 9.10 64bit OS. Note that for a developer, the ability to run a local sandbox on his/her own notebook, with Internet connection optional, really brings out a level of productivity, comfort and convenience that using a live remote cloud storage system never provides.
One of the two KVM guest acts as the ZCloud Sandbox server, and another one acts as a client host. Each KVM guest has 512MB RAM, 12 GB disk storage, and two 2.53GHz VCPUs. Hardware-wise, the notebook has an Intel Core 2 Duo P8700 CPU @ 2.53Ghz, 8GB RAM, and a 80 GB Intel X-25M G2 SSD.
Shown below is a screen shot of the KVM Virtual Machine Monitor. While the test was running, the ZCloud Sandbox server (server0) had a negligible load. The client (client0), however, was busy fetching and throwing data objects at the server.