ioTurbine: improving the performance of virtualization systems
ioTurbine is an application designed for transparent caching of transferred data from SAN or NFS storage systems in virtual environments. ioTurbine caches host data, providing low latency. Using ioTurbine software and Fusion-io cards, users can get cache terababytes based on flash memory technology, significantly increasing the performance of virtual environments, eliminating disk subsystem bottlenecks, increasing the number of virtual machines located on one server and thereby minimizing storage infrastructure costs.
Although virtualization allows organizations to use and manage infrastructure much more efficiently, it places extremely high demands on storage systems, both with sequential and random access to storage data. But just random data operations are the biggest problem when working with any type of storage system. And it is to eliminate this drawback to spend the largest part of the funds invested in storage systems. As a result, most of the funds saved by operating a virtual work environment are spent on the purchase of more expensive and more productive storage systems.
ioTurbine unleashes the potential of VMware, turning ioDrive cards into a powerful and easy-to-manage tool to improve storage performance in a virtual environment. Once installed, ioTurbine runs in the background as a component of the hypervisor and guest operating system. For optimal use of resources, ioTurbine dynamically balances I / O operations between virtual machines, and vMotion is supported as well as moving machines between hosts.
')
Due to the fact that ioTurbine works transparently for virtual machines, configuration changes can be made on the fly, without machine downtime. The administrator can configure the priority between virtual machines and applications, ensuring the greatest performance exactly where it is needed at the moment. Due to the unloading of the main data storage, the productivity of applications and the system as a whole is increased.
We decided to do some testing and find out what the percentage of growth in “parrots”, i.e. in virtual machines.
The IBM System x3650 M3 server was selected as the test bench in the following configuration:
The testing method is simple, the N-th number of virtual machines with the cache turned off was taken, the tests were run, then the cache for all the machines was turned on and the tests were run again.
As for the tests, there were 2 of them:
In fact, initially we wanted to ensure that the performance on virtual machines without a cache would simply go to 0 and see how many virtual machines after that could be added by connecting the cache. But testing and studying this system took a very large amount of time from us, so it was finally decided to conduct performance measurements, taking as many as 10 virtual machines.
Iometer
SQLIO Random Reads
Iops
MB / s
SQLIO Sequential Reads
Iops
MB / s
As we can see, in the more synthetical Iometer test, the difference is not so great after the launch of the second virtual machine, but with the more real SQLIO test that emulates the operation of the SQL server, the indicators become more joyful and only at the end, already on the 10th virtual machine their indicators become closer to each other, although in all other cases - work with the cache is 1.5-2 times faster than without it.
To my great regret, we did not have time to test with live migration of virtual machines, but perhaps we will return to this issue in one of our future articles.
What I want to say in the end about ioTurbine - if you are limited in the possibilities of expanding your server, ioTurbine + ioDrive is quite a good solution for increasing the performance of the disk subsystem.
To purchase a Fusion-io or request a Demo card, please contact Alexey Kotov, the head of the referral.
By KorP
Although virtualization allows organizations to use and manage infrastructure much more efficiently, it places extremely high demands on storage systems, both with sequential and random access to storage data. But just random data operations are the biggest problem when working with any type of storage system. And it is to eliminate this drawback to spend the largest part of the funds invested in storage systems. As a result, most of the funds saved by operating a virtual work environment are spent on the purchase of more expensive and more productive storage systems.
ioTurbine unleashes the potential of VMware, turning ioDrive cards into a powerful and easy-to-manage tool to improve storage performance in a virtual environment. Once installed, ioTurbine runs in the background as a component of the hypervisor and guest operating system. For optimal use of resources, ioTurbine dynamically balances I / O operations between virtual machines, and vMotion is supported as well as moving machines between hosts.
')
Due to the fact that ioTurbine works transparently for virtual machines, configuration changes can be made on the fly, without machine downtime. The administrator can configure the priority between virtual machines and applications, ensuring the greatest performance exactly where it is needed at the moment. Due to the unloading of the main data storage, the productivity of applications and the system as a whole is increased.
We decided to do some testing and find out what the percentage of growth in “parrots”, i.e. in virtual machines.
The IBM System x3650 M3 server was selected as the test bench in the following configuration:
- Xeon E5506 processor
- 48GB memory
- hard drives 8xSAS 146GB 10k
- ioDrive2 365Gb MLC
The testing method is simple, the N-th number of virtual machines with the cache turned off was taken, the tests were run, then the cache for all the machines was turned on and the tests were run again.
As for the tests, there were 2 of them:
- Iometer (70R / 30W)
- SQLIO Random Reads
- SQLIO Sequential Reads
In fact, initially we wanted to ensure that the performance on virtual machines without a cache would simply go to 0 and see how many virtual machines after that could be added by connecting the cache. But testing and studying this system took a very large amount of time from us, so it was finally decided to conduct performance measurements, taking as many as 10 virtual machines.
Iometer
SQLIO Random Reads
Iops
MB / s
SQLIO Sequential Reads
Iops
MB / s
As we can see, in the more synthetical Iometer test, the difference is not so great after the launch of the second virtual machine, but with the more real SQLIO test that emulates the operation of the SQL server, the indicators become more joyful and only at the end, already on the 10th virtual machine their indicators become closer to each other, although in all other cases - work with the cache is 1.5-2 times faster than without it.
To my great regret, we did not have time to test with live migration of virtual machines, but perhaps we will return to this issue in one of our future articles.
What I want to say in the end about ioTurbine - if you are limited in the possibilities of expanding your server, ioTurbine + ioDrive is quite a good solution for increasing the performance of the disk subsystem.
To purchase a Fusion-io or request a Demo card, please contact Alexey Kotov, the head of the referral.
By KorP
Source: https://habr.com/ru/post/189110/
All Articles