Caché 2015 scalability with Ivy Bridge-EX

    Recently, we talked about the latest release of the InterSystems Caché DBMS 2015.1 , which significantly improved performance, supported the second version of the XSL Transformations (XSLT) Version 2.0 XML document transformation language and the use of semaphores to synchronize processes in Caché and organize their interaction with each other. In addition, the new Caché release expands the use of high-availability scripting and supports the use of an external NGINX web server for the Caché application server. In this article, we will talk about the first of the listed improvements to our DBMS, namely, how Caché's performance improves when working on a new release on multiprocessor servers equipped with Intel Xeon E7 v2 processors (codenamed Ivy Bridge-EX).

    These processors, which are the most powerful in the Intel Xeon line and are designed for four or more processor systems, were released about a year ago. Compared with the first generation Xeon E7, they have twice the performance, built-in cache by 25%, the maximum amount of supported RAM, and the I / O channel bandwidth by four times. In addition, by utilizing crash protection technologies developed for Intel Itanium processors, Xeon E7 v2-based servers provide five-nine availability (99.999%). The Xeon E7 processor contains up to 15 processor cores that support two instruction streams and use a shared L3 cache of up to 37.5 MB and up to three QuickPath Interconnect (QPI) links. Four processor servers equipped with this processor

    In order to evaluate the scalability of Caché 2015.1 using Xeon E7 v2 processors, InterSystems tested the database while maintaining EPIC's Electronic Medical Records (EMR), which is used in a number of the largest American medical institutions. Tests showed that Caché 2015.1 in combination with Enterprise Cache Protocol (ECP) technology on Xeon E7 v2-based servers is capable of serving more than 21 million end-user database queries per second (we call this performance metric Global References per Second or GREF) . At the same time, the result that was achieved at Caché 2013.1 in combination with the Xeon E5 was improved more than three times. The choice for testing the EMR system is explained by the fact that medical institutions, along with financial institutions, retail and the public sector, are the main customers of InterSystems.

    Threefold productivity gains have been achieved with the following Caché 2015.1 scalability enhancements:

    • A new parallelization algorithm for certain applications, thanks to which, with an increase in the load and the number of users, the response time of applications remains at the same level;

    • Caché 2015.1 uses special commands that minimize delays when running applications on multicore systems associated with non-uniform access to data (NUMA, non-uniform memory access), therefore, when the number of processors installed in the servers increases, the application response time does not increase.

    How were the tests conducted?

    Tests were conducted using a copy of a real multi-terabyte database, which is used by the EPIC EMR system, which serves the Sanford Health hospital network. Sanford Health is one of the largest medical organizations in the USA, which owns 43 hospitals with 1,400 doctors, and provides treatment for more than 26 thousand patients.

    Performance experts Intersystems and Epic performed a series of tests on a multi-terabyte copy of the Sanford Health database to determine Caché 2015.1 performance measured in GREF, and also tested the same database with the previous Caché 2013.1 release. To generate the load created by the database queries from end users, special modeling tools developed by Epic were used. At the same time, the number of end users accessing the database gradually increased. In the tests, we measured both the GREF performance indicators and the delay value, i.e. how fast the database can handle a series of complex queries.

    To test Caché 2015.1, we used a four-processor server with Intel Xeon E7-4890 v2 processors with a clock frequency of 2.8 GHz, and Caché 2013.1 worked on a two-socket server with Intel Xeon E5-2680 2.7 GHz. Both servers ran Red Hat Enterprise Linux and VMware vSphere; they possessed the necessary resources of computing power, RAM and disks, so that these indicators do not limit the scalability of systems or slow down the response of the database to queries. A solid-state array was used as the storage system.


    As you can see from the graph, Caché 2015.1 on a server with Intel Xeon E7-4890 v2 in tests showed three times higher performance than Caché 2013.1 on a server with Intel Xeon E5-2680. The graph also demonstrates that with an increase in the number of queries to the database, delays when using the new Caché release grow much more slowly. It is worth noting that testing with more than 22 million GREF was not conducted due to limitations in the hardware configuration of the server, and not to achieve the performance ceiling of Caché 2015.1 itself.


    Testing demonstrated significantly improved scalability of the new Caché release, provided both by improvements in the parallelization mechanism of the database itself and by taking advantage of the new multi-core architecture of Intel Xeon E7 v2 processors.

    Thank you for your attention, we are ready to answer your questions.

    Also popular now: