Background

In recent years, as cloud computing has developed rapidly, a large number of Jiangsu Telecom business platforms have moved to the cloud, the number of virtual machines has risen sharply, and capacity and performance of the existing resource-pool compute and storage have become insufficient. The storage system has been planned by tier to meet differentiated business storage-performance needs, but although midrange storage is deployed, high-end and low-end storage are still lacking. Following the tiered-storage plan, China Telecom Jiangsu proposed object-based distributed storage for low-end use, so as to raise storage-space utilization and lower cost.

Requirements Analysis

With the spread of virtualization, China Telecom Jiangsu has fully turned to cloud computing to build its new IT systems. A large amount of business has already moved to virtual machines, and the back end also plans to use this cloud-storage procurement to build a large storage resource pool for unstructured data of Jiangsu Telecom business platforms, mainly massive small-file read/write such as CDRs, logs, security records, and backup data, plus some large-file read/write.

After long development, virtual-machine performance can meet many application needs, especially processor and memory virtualization, which is close to physical-host performance. However, input/output (I/O) device performance remains unsatisfactory.

First, although virtualization makes each virtual machine believe it has independent dedicated resources, the I/O performance of each storage system is limited. I/O contention among virtual machines limits actual I/O performance. Unlike CPU and memory allocation, I/O is limited by the principles of traditional mechanical disks, with inherently slower processing that struggles with load fluctuation. Burst I/O can affect service quality or even stall service.

Therefore, as a hybrid-cloud service provider, Jiangsu Telecom required the newly procured storage system to have the following characteristics:

(1) Robustness, providing sound isolation for virtual machines that share resources;

(2) Efficiency, using I/O resources effectively and minimizing response time;

(3) Scale-out, so that expansion can easily meet business growth for years to come;

(4) Fault tolerance, so that mechanical faults or partial data loss do not affect normal business and automatic fault recovery is possible;

(5) High cost-performance, built on standard rack servers, with an easy-to-use management UI;

(6) Support for PB-class massive storage.

Solution

After in-depth discussion, the customer had studied distributed storage in depth and believed it could resolve insufficient I/O performance, insufficient storage space, and high later expansion cost in the current IT system. On that basis, we recommended Sugon’s large-scale distributed storage system ParaStor, which has the following characteristics:

(1) Very high I/O bandwidth and IOPS;

(2) Built on standard x86 storage servers, so later expansion is convenient and cost-performance is high;

(3) N+M erasure-coding fault tolerance, easily handling data loss, disk faults, and similar issues;

(4) Easy scale-out to PB-class massive storage;

(5) A mature, stable storage file system recognized by a large number of users.

Through multiple exchanges, the customer understood ParaStor’s features and performance, conducted strict tests, compared it in detail with similar products, and ultimately accepted Sugon’s product.