Cisco Hyperflex for high-load DBMS
We continue the series of articles about Cisco Hyperflex. This time we will acquaint you with the work of Cisco Hyperflex in conditions of high-loaded Oracle and Microsoft SQL DBMS, and also compare the obtained figures with competitive solutions.
In addition, we continue to show the possibilities of Hyperflex in the regions of our country and we are pleased to invite you to visit the next demonstration of solutions, which this time will be held in the cities of Moscow and Krasnodar.
Moscow - May 28 Record by reference .
Krasnodar - June 5th. Record by reference .
')
Until recently, hyperconvergent solutions were not a very suitable solution for a DBMS, especially with a high load. However, thanks to the use of the UCS factory as a hardware platform for Cisco Hyperflex, which has proven its reliability and performance over 10 years, this situation has already changed.
Want to know more? Then welcome under cat.
At the moment there are two approaches to the organization of hyperconvergent solutions. The first approach is based on software-defined solutions that are delivered as software, and customers choose the equipment themselves. The second approach is based on turnkey solutions, that is, containing both software, hardware, and technical support. We at Cisco adhere to the second approach and deliver ready-made solutions to our customers, because only in this way can we ensure stable system behavior, high-quality technical support from one manufacturer and high performance.
It is the high system performance that is one of the key factors in deciding on the use of a particular product in mission-critical tasks.
Today, organizations typically place critical tasks on classic three-tier architecture solutions (storage> storage network> servers). At the same time, most organizations strive to simplify and cheapen the IT infrastructure, while not reducing its stability and performance. For this reason, more and more often customers are paying attention to hyper-convergent solutions.
As part of this article, we will talk about the latest tests (February 2019), which were performed by an independent ESG laboratory (Enterprise Strategy Group). During testing, the work of highly loaded Oracle and MS SQL DBMS (OLTP tests) was emulated, which is one of the most critical components of the IT infrastructure in a real productive environment.
This load was performed on three solutions: Cisco Hyperflex, as well as two software-defined solutions that were installed on the same servers that are used in Hyperflex, that is, on Cisco UCS servers.
The vendor A system does not use the cache, since the cache configuration is not supported by the solution developer. For this reason, discs were used to store a larger volume.
OLTP tests were performed with four virtual machines and a 3.2TB working dataset. Before each test, each VM was filled with recorded data using the testing tool. This ensures that the test reads “real” data and writes it into existing blocks, rather than simply returning zero blocks or zero values ​​directly from memory. This happens when the data is not filled, so it was important to make sure that the test accurately reflects how the data is read and written in the application environment. Filling this large work package took a long time, but in our opinion it is a productive investment of time, because it allows you to get more accurate performance data.
Testing was conducted using the HCI Bench tool (based on Oracle Vdbench) and I / O profiles designed to emulate complex critical OLTP workloads using Oracle and SQL Server backends. The block sizes were assigned in accordance with the emulated applications with 100% random data access (full random).
The first was an OLTP test designed to emulate an Oracle environment. Vdbench was used to create a workload with different read / write ratios. The test was conducted on four virtual machines. During the four-hour test, HyperFlex managed to show over 420,000 IOPS with a delay of just 4.4 milliseconds. Software solutions A and B were able to show only 238,000 and 251,000 IOPS, respectively.
The level of delays in different systems was about the same, with the exception of the write delay for vendor B, which averaged 26.49 ms, with very good read delays of 2.9 ms. Compression and deduplication were active in all systems.
Then an OLTP workload was considered for emulating a Microsoft SQL Server DBMS.
As a result of this test, the Cisco HyperFlex cluster is approximately twice as large as both competitors A and B. 490,000 IOPS from Cisco against 200,000 and 260,000 from manufacturers A and B.
The result of delays in Cisco HyperFlex was not very different from the Oracle test, that is, it was at a good level of 4.4 ms. At the same time, manufacturers A and B showed significantly worse results than in the test for Oracle. The only positive point for a competitive solution B is a consistently low level of delays when reading at 2.9 ms, in all other indicators Hyperflex outpaced competitive solutions twice and more.
Testing conducted by the independent ESG laboratory not only once again confirmed the decent performance level of the Cisco Hyperflex solution, but also proved that hyperconverged systems are already ready for general use in mission-critical tasks.
Hyper-converged systems have long been considered more suitable for non-critical loads. In 2016, ESG conducted a survey among large companies. They were asked why they preferred traditional infrastructure, rather than hyperconvergent. 54% of respondents answered that the reason is performance.
Fast forward to 2018. The picture has changed: a re-survey of ESG has already revealed only 24% of respondents, who still believe that traditional approaches are still better in terms of performance.
When the evolution of technology changes the criteria for choosing an industry solution, there is often a mismatch between what customers want and what they can get. Manufacturers who can see what is missing and fill this void - get an advantage. Cisco offers a hyper-convergent solution that provides simplicity and cost-effectiveness, as well as consistently high performance, which is not enough and which customers need for mission-critical workloads.
Cisco is moving forward steadily in the field of hyperconvergent systems, which is confirmed not only by the excellent characteristics of the Cisco Hyperflex solution, but also by its market presence. Therefore, in the fall of 2018, Cisco deservedly entered the HCI market leader group according to Gartner.
Already, you can see that Hyperflex is an excellent solution for the most complex and demanding business problems by visiting our demonstrations, which will be held in the cities of Moscow and Krasnodar.
Moscow - May 28 Record by reference .
Krasnodar - June 5th. Record by reference .
In addition, we continue to show the possibilities of Hyperflex in the regions of our country and we are pleased to invite you to visit the next demonstration of solutions, which this time will be held in the cities of Moscow and Krasnodar.
Moscow - May 28 Record by reference .
Krasnodar - June 5th. Record by reference .
')
Until recently, hyperconvergent solutions were not a very suitable solution for a DBMS, especially with a high load. However, thanks to the use of the UCS factory as a hardware platform for Cisco Hyperflex, which has proven its reliability and performance over 10 years, this situation has already changed.
Want to know more? Then welcome under cat.
Introduction
At the moment there are two approaches to the organization of hyperconvergent solutions. The first approach is based on software-defined solutions that are delivered as software, and customers choose the equipment themselves. The second approach is based on turnkey solutions, that is, containing both software, hardware, and technical support. We at Cisco adhere to the second approach and deliver ready-made solutions to our customers, because only in this way can we ensure stable system behavior, high-quality technical support from one manufacturer and high performance.
It is the high system performance that is one of the key factors in deciding on the use of a particular product in mission-critical tasks.
Today, organizations typically place critical tasks on classic three-tier architecture solutions (storage> storage network> servers). At the same time, most organizations strive to simplify and cheapen the IT infrastructure, while not reducing its stability and performance. For this reason, more and more often customers are paying attention to hyper-convergent solutions.
As part of this article, we will talk about the latest tests (February 2019), which were performed by an independent ESG laboratory (Enterprise Strategy Group). During testing, the work of highly loaded Oracle and MS SQL DBMS (OLTP tests) was emulated, which is one of the most critical components of the IT infrastructure in a real productive environment.
This load was performed on three solutions: Cisco Hyperflex, as well as two software-defined solutions that were installed on the same servers that are used in Hyperflex, that is, on Cisco UCS servers.
Test configurations
The vendor A system does not use the cache, since the cache configuration is not supported by the solution developer. For this reason, discs were used to store a larger volume.
Testing Methodology
OLTP tests were performed with four virtual machines and a 3.2TB working dataset. Before each test, each VM was filled with recorded data using the testing tool. This ensures that the test reads “real” data and writes it into existing blocks, rather than simply returning zero blocks or zero values ​​directly from memory. This happens when the data is not filled, so it was important to make sure that the test accurately reflects how the data is read and written in the application environment. Filling this large work package took a long time, but in our opinion it is a productive investment of time, because it allows you to get more accurate performance data.
Testing was conducted using the HCI Bench tool (based on Oracle Vdbench) and I / O profiles designed to emulate complex critical OLTP workloads using Oracle and SQL Server backends. The block sizes were assigned in accordance with the emulated applications with 100% random data access (full random).
Oracle Database workload
The first was an OLTP test designed to emulate an Oracle environment. Vdbench was used to create a workload with different read / write ratios. The test was conducted on four virtual machines. During the four-hour test, HyperFlex managed to show over 420,000 IOPS with a delay of just 4.4 milliseconds. Software solutions A and B were able to show only 238,000 and 251,000 IOPS, respectively.
The level of delays in different systems was about the same, with the exception of the write delay for vendor B, which averaged 26.49 ms, with very good read delays of 2.9 ms. Compression and deduplication were active in all systems.
Microsoft SQL Server Workload
Then an OLTP workload was considered for emulating a Microsoft SQL Server DBMS.
As a result of this test, the Cisco HyperFlex cluster is approximately twice as large as both competitors A and B. 490,000 IOPS from Cisco against 200,000 and 260,000 from manufacturers A and B.
The result of delays in Cisco HyperFlex was not very different from the Oracle test, that is, it was at a good level of 4.4 ms. At the same time, manufacturers A and B showed significantly worse results than in the test for Oracle. The only positive point for a competitive solution B is a consistently low level of delays when reading at 2.9 ms, in all other indicators Hyperflex outpaced competitive solutions twice and more.
findings
Testing conducted by the independent ESG laboratory not only once again confirmed the decent performance level of the Cisco Hyperflex solution, but also proved that hyperconverged systems are already ready for general use in mission-critical tasks.
Hyper-converged systems have long been considered more suitable for non-critical loads. In 2016, ESG conducted a survey among large companies. They were asked why they preferred traditional infrastructure, rather than hyperconvergent. 54% of respondents answered that the reason is performance.
Fast forward to 2018. The picture has changed: a re-survey of ESG has already revealed only 24% of respondents, who still believe that traditional approaches are still better in terms of performance.
When the evolution of technology changes the criteria for choosing an industry solution, there is often a mismatch between what customers want and what they can get. Manufacturers who can see what is missing and fill this void - get an advantage. Cisco offers a hyper-convergent solution that provides simplicity and cost-effectiveness, as well as consistently high performance, which is not enough and which customers need for mission-critical workloads.
Cisco is moving forward steadily in the field of hyperconvergent systems, which is confirmed not only by the excellent characteristics of the Cisco Hyperflex solution, but also by its market presence. Therefore, in the fall of 2018, Cisco deservedly entered the HCI market leader group according to Gartner.
Already, you can see that Hyperflex is an excellent solution for the most complex and demanding business problems by visiting our demonstrations, which will be held in the cities of Moscow and Krasnodar.
Moscow - May 28 Record by reference .
Krasnodar - June 5th. Record by reference .
Source: https://habr.com/ru/post/452602/
All Articles