Why caching is an important element in enhancing SmartCloud input/output performance

To understand the transient storage performance in the http://www.aliyun.com/zixun/aggregation/13696.html ">smartcloud environment, we set up a benchmark test using Vdbench. Vdbench is an I/O workload builder that validates data integrity and measures the performance of storage for direct attach and network connectivity. It is a free tool that is easy to use and is often used for testing and benchmarking.

We conclude that caching is an important element in I/O benchmarking. This article describes the steps we set up to benchmark and shows the results. But first we will briefly introduce the following products and tools that we use.

Our benchmark settings

Starting with SmartCloud, this is a public cloud service built on the basis of a kernel based virtual machine (KVM), a comprehensive virtualization solution for linux® on the x86 architecture. SmartCloud is a complete infrastructure as a service (IaaS) product. It can be used as a foundation for building a platform, a service (PaaS), and a software-service (SaaS) solution.

Key features of SmartCloud include:

Self-service configuration Virtual server machine and virtual storage space (persistent storage) network function automatically configures the resource with a paid settlement policy the open API can be used to develop scripts and software to enhance the automation VM size covering from one virtual CPU and 2GB memory to 16 CPUs and 32GB RAM. Both Windows® and Linux operating systems can be deployed, including Red Hat Enterprise Linux and SUSE Linux.

SmartCloud provides 3 types of storage:

Short-term storage is associated with a VM being rationed. Persistent storage is a network-attached storage that is accessed by dynamically attaching to an active instance/separating from that instance. Object storage is jointly provided with Nirvanix and can be configured as an on-demand storage solution for unstructured data.

In our benchmark, we'll focus on short storage.

We use Vdbench to test the original disk and file system. It has a WEB user interface to display detailed performance reports. Vdbench was developed by the Henk Vandenbergh of Sun Microsystems, formerly known as StorageTek. Vdbench is written using java®. It has been tested on Solaris, Windows, HP-UX, AIX, Linux, Mac OS X, Zlinux, and native VMware.

More information for temporary storage

Ephemeral storage is created when configuring a VM. Its lifecycle is directly related to the instance to which it is bound. Ephemeral storage is created from a local disk located within a node. It has significant throughput changes based on the operations being performed on the shared infrastructure.

For configuration transient storage is not less than 60GB, for the maximum configured VM, up to 2048GB. If it is not necessary to use the full instance store, you can configure the virtual machine with a minimum amount of temporary storage (GB). Configuring minimum storage reduces the configuration time for large virtual machine types. When an instance is deleted, the virtual machine instance store is erased.

Figure 1. Configure a transient storage for a VM

Conversely, to store data for longer periods of time, you can use persistent storage blocks. Persistent storage has no lifetime to bind to a VM compared to a transient storage, which is calculated independently. Persistent storage can be dynamically attached to or detached from a VM because it is a network-attached storage (NAS) raw disk that must be formatted and loaded from the guest, as shown below.

Figure 2. Persistent storage through network-attached VMS

In our benchmark test, we created a transient storage by configuring a VM with Red Hat Linux. It resembles the following figure:

Figure 3. Configuring VMS for transient storage using Red Hat Linux

Set up benchmark tests

You should have installed and running SmartCloud. Next, you must assign a virtual block device and a ext3 file system, and then install Vdbench.