It is that time of year to use those precious vacation days to seek out secluded beaches and scenic mountains before the sunny weather is gone. But how long are you willing to travel to reach your destination? Proximity does affect your overall travel time and how many days you are vacationing. And then, what to do with all your luggage-can it all fit on the plane, so you get to the beach as soon as you land?
The same theory applies when you consider memory options for your server. The closer you store data next to your CPU, the faster it can travel and become useful to your organization. Look at it as carry-on luggage vs checked luggage but without the size constraints. Dell PowerEdge recently added a new storage-class memory for select servers designed to accelerate databases, transactions, data analytics and virtualization. Intel® Optane™ DC Persistent Memory offers the speed of memory, economical storage and data persistence.
With the close proximity of storage-class memory to the CPU, performance can increase since data does not have to make that round trip to storage for every transaction. When we compared NVMe drives to Intel® Optane™ DC persistent memory in a Microsoft SQL 2019 preview environment, we saw up to 2.7x the transactions per second![1]
Intel® Optane™ DC Persistent Memory is currently available in three capacities: 128GB, 256GB, and 512GB. This is significantly larger than typical DRAM options. This larger capacity is particularly relevant when running workloads like SAP HANA that have extensive data. In fact, the PowerEdge R940 achieved an initial 11.7B records in a recent SAP BW edition for SAP HANA benchmark with persistent memory.[2]
Plus, the larger memory capacity can offer greater cost-savings. A PowerEdge R940 server running SAP Business Warehouse with Intel® Optane™ DC Persistent Memory and DRAM achieves similar performance compared to a DRAM-only configuration during data loads with up to 24% lower memory cost.[3]
Intel® Optane™ DC Persistent Memory runs in two different modes: Memory Mode, where the memory exists “above” the DRAM and acts as a cache, and App Direct Mode, where the DRAM and persistent memory are pooled together to maximize the total capacity. Any application (like virtualization or databases) can take advantage of data center persistent memory modules (DCPMM) in Memory Mode with a compatible operating system. You can unlock more performance as well as persistency when using an application that supports App Direct Mode, but not every workload is ideal for this kind of configuration. DCPMMs are used in conjunction with RDIMMs or LRDIMMs and a maximum number of 6 DCPMMs can be used per CPU. The chart below depicts features and capacity for the two memory modes.
Trait | Memory Mode | App Direct Mode |
Application support | Any application | Application must state that it supports “App Direct Mode” |
DRAM | Used as cache and is not available as system memory | Both DCPMM and DRAM are available as system memory |
Maximum Memory Capacity | R640, R740, R740xd, MX740c two-socket PowerEdge servers: 6.14TB (DCPMMs)
R840, R940, R940xa, MX840c four-socket PowerEdge servers: 12.28TB (DCPMMs) |
R640, R740, R740xd, MX740c two-socket PowerEdge servers: 6.14TB (7.68TB total DCPMMs + LRDIMMs)
R840, R940, R940xa, MX840c four-socket PowerEdge servers: 12.28TB (15.36TB total DCPMMs + LRDIMMs) |
Persistence | No | Yes |
I hope you travel lightly and enjoy these last few official days of summer. When you return to work, be sure to join the conversation @Dellservers on Twitter and let us know how persistent memory is accelerating your data-heavy workloads.
[1] Based on Dell internal testing in March 2019 comparing a PowerEdge R740xd with 2nd Generation Intel® Xeon® Scalable processors and 2 x 1.6TB Mixed Use NVMes using Microsoft SQL Server 2019 CTP 2.3, VMware ESXi 6.7U1 and a RHEL 7.6GA vs the same R740xd configuration except for substituting the NVMes for 12 x 256GB DIMMs of Intel® Optane™DC Persistent Memory. Workload used is derived from TPC-C to SQL Server 2019 CTP 2.3 performance. The results obtained with the derived workload are not comparable to published TPC-C results. Actual results will vary.
[2] Based on results of the Dell PowerEdge R940 on the SAP BW edition for SAP HANA benchmark Version 2; Initial Records: 11.7 billion, Phase 1: Data Load 30,856 seconds, Phase 2: Query Executions per Hour 2,507 with 892.01 billion Records Selected, Phase 3: Runtime 122 seconds, Benchmark: Version 3, Environment: Bare-Metal, SAP Technology Release: SAP NetWeaver 7.50, Operating System: RHEL 7.6, Database Release: SAP HANA 2.0, Configuration: Scale-Up. 1 Node with 4 x Intel Xeon Platinum 8280L Processors (2.7 GHz, 112 cores, 224 threads), 1,536 GB of DRAM and 3,072GB of persistent memory. Certification number: 2019041. Results as of June 2019. For more details see http://sap.com/benchmarks. Actual results will vary.
[3] Based on Internal Dell testing and list pricing comparison August 2019. Tests run on PowerEdge R940 comparing a configuration with 24 x 128GB DCPMM + 24 x128GB LRDIMM (App Direct Mode) vs a configuration with 48 x 128GB LRDIMM. The “load time” for 2.6B records was 41,202 seconds with DCPMMs +LRDIMMs vs. 40,488 seconds with LRDIMMs only. The R940 base configuration included a PERC H740P 8GB cache, RAID5, 12 x 800GB drives, SLES 15, HANA 2.0 Revision 41. Results will vary. Price analysis based on current list pricing on 8/30/19, subject to change.