Openstack kills VMWare (2)

The last article "Openstack kills VMWare (1)" generally talked about some comparisons between using the open source version of openstack and vmware. Note that I am talking about the open source version of openstack. As for the various commercial versions of openstack, they are not in the scope of my discussion. I didn’t pay me a salary to advertise for them. Haha, the preliminary conclusion is that through the efforts of the openstack open source community, the open source version of openstack now has the conditions to actually land on small and medium-sized enterprises and kill vmware, so that the masses of people can use it. Go to your own open source openstack private cloud, and continue to talk about it below.

Now OpenStack is used by many real-world production cases using distributed storage ceph for each component, and today we analyze the comparison between OpenStack and VMware primarily from the storage perspective. First of all, the relationship between OpenStack and Ceph, Ceph is the mainstream back-end storage of OpenStack, and OpenStack Brothers, to provide unified shared storage services for OpenStack. Following a brief introduction to Ceph distributed storage, Ceph is a rapidly emerging open source distributed storage system that has several features:

1. Multi-copy feature, which is similar to the redundancy of the RAID1 features to ensure data security, can be 1 copies of 2 copies of 3 copies of multiple copies;

2. Copy the cow feature when writing, that is, when the data is written to occupy the storage space, with this feature, the use of space efficiency than traditional storage greatly improved;

3. Data stripe feature, that is, io concurrency read and write characteristics, so as to solve a reading efficiency problem, that is, the more disk, read and write IO efficiency higher;

4.thin provisioning feature, i.e. on-demand space, sparse sparse file similar to Linux file system;

I try to describe in the most concise language, many other professional concepts and descriptions are interested, you can refer to the relevant information on the Internet. The above features are a revolutionary thing for traditional storage and are now described in terms of their own understanding.

When I first contacted Ceph, there were some doubts, such as: for a small-scale application scenario, Ceph is not overqualified, if using multiple copies, such as the general use of 3 copies, the actual capacity of only one-third, the space waste is very serious, how to achieve redundancy and efficiency balance; How to ensure that data is secured in the event of a single physical machine failure with multiple disks, for example, the use and maintenance of ceph will be cumbersome, hard disk failure replacement is not very troublesome and so on. Now, after a certain use of practice, my doubts have vanished.

first of all, the most practical issues, redundancy and space use efficiency issues, I have done my own testing, in ensuring data security, ceph cluster data pool with 3 copies of the case, OpenStack hatching 10 20G size RAW format virtual machine, The actual use of ceph storage space is more than 30 g, according to traditional storage, if using the raw format of the image, Occupy space is real deal, 10 20G size of virtual machine, then occupy space is 200G, so, for the multi-copy of Ceph to ensure data security and "waste" out of space, it is easy to be "Savings" came back, so Ceph's space use efficiency is very high. Where it is how to achieve, too professional I do not want to describe more than I also have some weakness, you can refer to my other blog "OpenStack uses ceph storage, what does Ceph do? " ".

Again, the practice of data redundancy. Because Ceph is a storage target scattered in multiple PG, multiple PG is distributed in the OSD, an OSD corresponding to a disk, if the server has more than one disk failure, so that all the disks and the OSD on this server is not available, then the data is not lost, after the actual verification, And will not be lost, ceph through the crush (controlled Replication under scalable Hashing) algorithm multi-level structure, with the concept of data "bucket", when allocating PG will avoid 2 copies in the same "bucket", for example, The first level of the bucket is the OSD (disk), the second level is host (host), the third level is chassis (frame), the 4th level is rack (rack), and so on, and so on, even you can define the new "bucket" level, but in general, with these common level is enough.

again, the use of maintenance is not a very complex problem. Ceph, deployed via Openstack Kolla, is implemented via Docker containers and can be easily implemented if a new disk OSD is required and the failed disk OSD is removed, which can be referenced in the two articles: "Openstack Kolla Deploy Ceph","Openstack ceph replacement failed disk ".

It seems like I haven't mentioned anything about VMware, and now it's a comparison of storage between VMware and OpenStack using Ceph.

1. Comparison of virtual machine redundancy features. OpenStack has a soft spot relative to VMware when it comes to production: The redundancy features of virtual machines are not well resolved, and since the combination with Ceph, the problem has been solved and costs almost 0. Why do you say that, VMware uses the storage vmotion feature of the virtual machine, need to purchase its enterprise version above this feature, in addition, to achieve all the virtual machine redundancy related functions, whether it is ha or vmostion, or ft, there is a precondition, Is that you must have shared storage, and configure the disk array as a shared storage everyone knows it's a small investment (except to set up a server to play as a shared storage). OpenStack uses ceph differently, does not require a complex configuration, does not require a centralized disk array, because CEPH itself is inherently distributed storage, natural is redundant features, so OpenStack uses Ceph storage, can easily realize the virtual machine thermal migration, That is, the virtual machine is not powered off from one physical server to another physical server, it is also easy to implement the redundancy of the virtual machine (evacuate) function, a physical server will be all the virtual machine to migrate to another server.

2. Cost-effective comparison of disk IO. The disk IO efficiency of VMware virtual machines relies on the physical hardware itself, such as the use of RAID0, data stripe, can improve parallel read and write speed, such as can be used by the SSD SSD, relying on centralized disk array of high-speed storage and so on. and the virtual machine disk IO efficiency of OpenStack using distributed storage ceph with distributed storage inherent distributed features, the more disk, parallel IO read and write efficiency, the same can be achieved by using SSD SSD to achieve high-speed storage pool, to meet different levels of storage needs, For example, a storage pool of generic disks is used for object storage, and a storage pool consisting of SSD SSDs is used for the image of Nova. What is the difference between the two, right, cost, is cost, ceph can not need disk array, all use the server itself, complete the construction of distributed storage, flexible and inexpensive.

3. Extensibility comparisons. Not to mention this aspect, VMware directly white flag. VMware expansion to the main bottleneck at the end of the storage, you must increase the capacity and performance of the centralized disk array, to expand to the limit of the disk array it uses, only to add new disk array for capacity expansion, and everyone knows, IT infrastructure is the most expensive, right, disk array, is the disk array. The use of Ceph is different, the storage capacity needs to be extended, directly on the physical server with the disk OSD, directly in the OpenStack cluster to add storage node host, and increasingly higher, Because the feature of distributed storage is the more OSD (object storage device), the higher the IO performance, and for Ceph, scalability is what it excels at.

This is where you write where you want to go, write it today, and write it back when you think of something new.

Openstack kills VMWare (2)