Our storage based on Tier 2 (and up) is a unique solution with the ability to reallocate disk space and volume without interrupting business processes, built-in mechanism for snapshots and backups.
It works using distributed stotage nodes, providing unprecedented performance of applications processing the information, such as databases.
Performance of processing data
In a lab environment we have achieved performance of one million IOPS to a single OCI instance with 32 GB RAM and 16 VCPU during random read test, and 3 Gb / s of continuous transfer for reading and saving.
To perform a meaningful test, FIO software was used with the following parameters:
fio --filename=/dev/sda --direct=1 --rw=randread --bs=4k --size=50G --numjobs=64 --iodepth=32 --runtime=120 --group_reporting --name=file1 --ioengine=libaio --thread --gtod_reduce=1 --use_os_rand=1
This means a fully random read test using blocks of 4 KB performed for 64 concurrent processes in the mode ignoring local cache of the operating system.
Similar results can be achieved during the OCI instance test, provided with Windows Server software using IOMeter.
The chart above shows a comparison between storage limited with regards to the IO operations to 120 000 IOPS, classic RAID and other software solutions. Storage Tier 2 (and up) is available to any customer who wants to exploit the true potential of the Oktawave cloud while maintaining the standard tariff plans.
Requirements Microsoft Exchange and data processing
To visualize the exceptional performance of our storage based on the Tier 2 (and up), we compared the requirements of an MS Exchange individual mailbox with the previous test. Here are the results.
As you can see, a single storage volume limited with regards to the IO operations to 120 000 IOPS is able to store 135 000 accounts of Exchange type. Note that in Oktawave there are no limits on the number of volumes. This means that anyone can use an instance of two or three sorage volumes to maintain a substantial database of mailboxes.
Load of the instance and competition
Our test scenario assumed operation of thousands of users who generate 10 thousand queries using the GET method. We could not test the 100 thousand.queries, since both Amazon and Rackspace failed to deal with this number of concurrent queries. In this test we used an instance with 16 GB of RAM and 16 VCPU. Apache WWW server was installed with no add-ons.
We can see that the instances limited with regards to the IO operations to 120 000 IOPS can support over 14 thousand. queries per second while maintaining the avarage time to service one request at 0.07 s.