KVM virtualization open-source high availability solution (5) Summary

 

Back to the origin of the question, why do we want to use an open-source solution?

Our goal is to find the best cost-effective solution that can solve our problems and be well controlled,

We hope that the open-source solution can achieve the following objectives:

1. High Reliability:

1) Good redundancy for disk, memory, Nic, CPU and even server faults;

2) Single Point or even multiple points of failure are allowed;

2. High Stability:

1) easy to build;

2) The software system is stable and does not need to be stuck in various bugs all day long;

3) The software system is capable of self-healing, automatic repair or quick and simple repair;

3. Good performance:

1) read/write iops can achieve or even surpass the performance of a single machine;

2) read/write iops can achieve the performance required by actual production;

Comparison of Reliability:

The distributed file systems listed above have their own unique features in terms of reliability. drbd is actually a dual-host hot backup. mfs allows storage node faults, but the metadata server has a single point of failure, although it can be restored through the log server, the file system stops service during the recovery process. glusterfs and sheepdog are network-mounted structures and are born with no separate faults. All of them are reliability-oriented, we should consider using glusterfs and sheepdog.

Comparison of stability:

Drbd, mfs, and glusterfs are all mature file systems. They have established and maintained ready-made documents and are widely used in the production environment. sheepdog is still a test type, considering its stability, it is not recommended in the production environment.

Performance Comparison:

After a simple test, the performance of drbd is slightly worse than that of a single machine. mfs, glusterfs, and sheepdog share a common feature, that is, poor write capability and good performance. After testing, to improve the write capability, the number of disks on a single storage node is critical, and the performance of network bandwidth is critical.

Final Conclusion

In terms of reliability, stability, and performance, the most suitable File System for kvm Virtualization in the production environment is glusterfs. A year later, sheepdog may be the most suitable!

Price comparison tables of various file systems

File System	Reliability: 5 points out)	Stability: 5 points out)	Performance 5 points out)	Total score of 15 points)
Drbd	4	4	3	11
Mfs	2	3	4	9
Glusterfs	4	4	4	12
Sheepdog	2	2	3	7

This article is from the "xiaoli110 blog". For more information, contact the author!