This site uses cookies. By continuing to browse the site you are agreeing to our use of cookies. Read our privacy policy>

Reminder

To have a better experience, please upgrade your IE browser.

upgrade

KunLun SAP HANA Appliance: Scale-Up First

Total: 0 products

  • Choose product to compare

Cancel
Compare with CISCO
Compare with H3C
  • Choose product to compare

Cancel
Compare with CISCO
Compare with H3C
  • Choose product to compare

Cancel
Compare with CISCO
Compare with H3C
  • Choose product to compare

Cancel
Compare with CISCO
Compare with H3C
MENU

KunLun SAP HANA Appliance: Scale-Up First

By Dr. Zheng Wei, Huawei IT Technology Planning Dept.

Professor Hasso Plattner, one of the founders of SAP in 1972, is a technology zealot. In 1998, he personally funded the establishment of the Hasso Plattner Institute (HPI) with a commitment to promote academic research in software system engineering and industrial applications. For nearly two decades, HPI’s biggest achievement has been the development and successful application of SAP HANA, a High-performance Analytic Appliance that uses in-memory database technology to enable the processing of massive amounts of real-time data in a short time.

In 2009, Dr. Plattner, representing HPI, published paper titled A Common Database Approach for OLTP and OLAP Using an In-Memory Column Database at the SIGMOD conference in Providence, Rhode Island. The core premise of the paper was the use of in-memory databases and columnar storage as the enablers of hybrid Online Transaction Processing and Online Analytical Processing (OLTP/OLAP) applications which, in turn, are the fundamental building blocks behind the development of the SAP HANA platform. Later, Dr. Plattner led the effort by HPI and SAP to successfully launch the SAP HANA in-memory database products, to which the market quickly and enthusiastically responded. As of 2017, the full spectrum of SAP applications has been updated to support the SAP HANA.

What are the Unique Advantages of SAP HANA?

These are the main technical features of SAP HANA:

• Keeps data as close as possible to computing resources, with database data residing entirely in dynamic memory. This avoids frequent data transfers from hard disk drives or other storage devices to and from memory. Except for persistence operations, the CPU avoids interactions with the storage system whenever possible.

• Uses column-based storage to improve the performance of OLAP services, which mainly involve ad-hoc queries, real-time reports, as well as improvements in compression efficiency.

• In the delta area of memory, data is stored in a mixed format of rows and columns to ensure OLTP performance.

So, how does SAP HANA keep data as close as possible to computing resources to support higher data computing performance, while at the same time adhering to the server ‘Scale-up First’ rule?

Technically speaking, the Von Neumann architecture of a computer has resulted in hierarchical storage, which is also called the ‘memory hierarchy.’ It has been widely accepted that in-memory data processing delivers much better performance than data processing based on any kind of external storage device. However, the notion that in-memory computing supports OLAP/OLTP hybrid applications without compromising performance has long been challenged. Dr. Plattner explained that, after analyzing the real-world application scenarios of SAP customers, he found that the data read/write behaviors of OLTP and OLAP workloads are not drastically different — OLTP applications are data read-dominant, while OLAP applications have similarly frequent data writes. Based on this statistical pattern, Dr. Plattner believes that in-memory database products are more than suitable for high-performance OLTP/OLAP hybrid applications.

Since the birth of SAP HANA, its market performance has been no less than brilliant due to its outstanding performance, especially in OLAP scenarios. SAP HANA has been highly praised by customers and has been rapidly gaining market share. Here is an example:

After an enterprise replaced a legacy database with the Huawei SAP HANA appliance, measurements of real-life data showed that average server response times were slashed from over 800 milliseconds to below 370 milliseconds. Customer response times were also reduced by nearly 100x for services such as remittance notifications via SMS queries. There are many such real-world cases that demonstrate an overall high level of customer satisfaction.

Figure 1: Characteristics of OLTP and OLAP workloads as identified by SAP

Objectively, memory-based databases have an absolute advantage over storage-based databases that run on the internal hard drives or external storage arrays. This is because they are not, technologically, the same generation of products. The current 64-byte local memory access latency for Intel® processors (Xeon E7 v4) is 110 nanoseconds, while the latency of reading 4 KB data from local SAS SSD disks is about 130 microseconds. As is clearly seen from the low latency measurements indicated above, in-memory access delivers on the promise of high performance.

HANA is the first successful in-memory database product and benefits from advances in technologies such as integrated semiconductor circuits. For example, DRAM density, DIMM capacity, and the number of CPU memory channels have all increased. The Intel® Brickland platform has achieved a memory capacity of over 1 TB on a single Broadwell CPU. The SAP HANA appliance is certified to operate up to 1 TB memory on a single Broadwell E7-8890 CPU.

Huawei KunLun: A Natural Fit with SAP HANA Software Features

Huawei KunLun has been certified for integration with a full range of SAP HANA appliance solutions. The KunLun SAP HANA appliance is based on the KunLun Mission Critical Server, which is engineered for business-critical workloads such as enterprise databases, decision support, and business processing. Supporting 4, 8, 12, 16, 20, or 32 Intel® processors with an ultra-large memory of up to 32 TB, the KunLun server is capable of efficiently processing massive amounts of data in large-scale in-memory computing systems. In fact, a single KunLun server will support a larger in-memory database than any existing 8-socket x86 server and, in so doing, greatly improve enterprise IT resource utilization and reduce management expenditures.

At present, the improvements in CPU performance have shifted from a sole reliance on faster clock frequencies to the use of multi-core parallel technologies. For example, each Intel® Broadwell processor supports up to 24 cores and 48 threads, and Huawei’s KunLun server provides a physical platform capable of supporting up to 768 cores and 1,536 threads. In accordance with the primary goal of SAP HANA appliances to keep data as close as possible to the CPU, KunLun servers are perfectly suited to supporting the SAP HANA software features for meeting customer demand for the ultimate in database performance.

For customers intent on selecting an in-memory appliance solution, SAP’s proposition is that the appliance be able to deliver best-in-class performance to customers. Based on this principle, SAP’s advice to customers is ‘scale-up first’ — which translates to a preference to meet memory capacity requirements in single-node configurations instead of cluster configurations. The implication of this principle is that cluster configurations cannot provide the best performance. In practical terms, if the memory required by a customer exceeds the capabilities of a single server (for example, a requirement for 100 TB cannot be met by any single x86 server), the customer system will have to be installed as a cluster. This is also when the experts in enterprise databases come into play with techniques such as dividing database tables into cluster nodes, or distributing data through the use of built-in algorithms.

For SAP HANA, differences in access modes lead to drastically different performance results between a single, high-performance node and a cluster of nodes that involve cross-node (cross-OS) I/O or Remote Direct Memory Access (RDMA) operations. Though not currently available, the roadmap is prepared for SAP HANA to support InfiniBand or RDMA over Converged Ethernet (RoCE). As detailed above, the latency when reading 64 bytes of data from the local memory of an Intel® E7 v4 processor is 110 nanoseconds. Using an InfiniBand interface supporting RDMA, the typical latency for reading 64 bytes from the memory of another node is 1.5 microseconds at a 40 Gbit/s InfiniBand Quadruple Data Rate (QDR). There is a 15x difference between the two modes. If a 10 GE interface is used (10 Gbit/s line rate and 1 Gbit/s effective bandwidth), there is at least a 60x difference between the two modes.

Figure 2: HANA single-node, big Non-Uniform Memory Access (NUMA) server vs. scale-out data access

There is an obvious necessity to avoid such cross-node data reads whenever possible, which requires databases or tables to be precisely divided. Though, in practical terms, making cross-node data reads is inevitable as the cross-node data accessed by enterprise applications is often correlated by nature. This makes the huge difference in the performance of single nodes versus clusters of nodes commonplace.

At present, the Huawei SAP HANA solution based on the KunLun Mission Critical Server supports up to 32 sockets with 32 TB of memory capacity on a single node. The 16 TB memory model is certified, the 20 TB model is planned for certification, and any specification above 20 TB will be certified on a case-by-case basis. KunLun single-node solutions, that experience shows meet the requirements for the majority of customers, support capacity expansion from 4 sockets to 32 sockets as may be needed to grow the customer’s business based on business need. KunLun provides a unique logical partitioning function that further divides resources. With the KunLun logical partitioning function, multiple SAP applications and HANA databases can be consolidated onto KunLun, meanwhile ensuring high performance and fault isolation. This feature is in the process of planning. For customers, the memory capacity and computing capability of the deployed Huawei SAP HANA appliance can be gradually expanded based on the actual service development requirements. In addition, the excellent performance of the single-node solution is always maintained. KunLun’s 32 TB memory capacity on a single node can meet most customers’ requirements. For hyper-scale applications requiring more than 32 TB of memory, Huawei can provide cluster solutions today and is well on track to providing next-generation KunLun-SAP HANA appliances capable of larger scales and higher performance.

Huawei: Best SAP Practitioner and Best Choice for Enterprise Digital Transformation

Used and trusted by many customers all over the world, Huawei’s KunLun SAP HANA solutions are serving more than 1,000 customers in over 40 countries and regions worldwide, including China, Europe, the Middle East, the Americas, and Africa. Current customers include China National Petroleum Corporation (CNPC), Spanish Petroleum Company (CEPSA), Hillarys Blinds in the U.K., Fonterra in New Zealand, and BYD Auto in China. CEPSA has praised KunLun for its outstanding performance and rock-solid reliability, and is planning to expand the use of the Huawei SAP HANA appliance to run its ERP core production system from a 13 TB Phase 1 deployment to 30 TB in Phase 2.

As the experience of delivering tangible benefits to enterprises with the SAP HANA system deepens, Huawei has risen to become the world’s best practitioner of SAP system infrastructure solutions and the best partner for enterprises undergoing their digital transformation.

ICT Insights on the go

Scan for mobile reading

""

Navigating Critical Enterprise Business Through Digital Transformation

Navigating Critical Enterprise Business Through Digital Transformation (03/2018)

Current catalogue

Comment

Industry Insight

Partner Testimony

Solution

Ecosystem