HPE Persistent Memory Microsoft Ignite 2017 Bret Gibbs HPE Persistent Memory Product Manager

What is Persistent Memory? “Fast storage at memory speeds with data persistence” What it provides to customers: Unlocks new levels of performance for business workloads Memory speeds without data volatility Complete hardware and software ecosystem solution How do we deliver it to customers: HPE NVDIMMs: “Right-sized capacity for smaller storage bottlenecks” 8GB NVDIMM (Gen9) 16GB NVDIMM (Gen10) HPE Scalable Persistent Memory: “Fastest performing persistent memory at terabyte scale” DL380 Gen10 with up to 1TB of capacity (Gen10) HPE Scalable Persistent Memory HPE 16GB NVDIMMs (Gen10) HPE 8GB NVDIMMs (Gen9)

HPE Scalable Persistent Memory Fastest performing persistent memory at terabyte scale Ideal for: Enabling in-memory compute with persistence, faster checkpoints and restores, HTAP, software-defined storage caching tiers and any workload that could benefit from low-latency DRAM-level performance Key Features: DRAM-level performance for fastest performing persistent memory Flexible capacity points with capacity up to 1TB Complete solution using DRAM for application performance, flash tier for persistence and backup power for moving data to persistent store Up to 20x reduction in database restart time preserving maximum uptime1 Up to 27x faster checkpoint operations enabling significantly faster business operations2 OS Support: Microsoft: WS2012 R2 (HPE driver) and WS2016 (inbox) Linux: RHEL 7.3 and SLES 12 SP2 HPE Scalable Persistent Memory is an integrated storage solution that runs at memory speeds with terabyte capacity unlocking new levels of performance for your business workloads. Key Messages: Fastest persistent memory offering on the market with DRAM-level performance and TB-scale capacity Supported on DL380 in Fall of 2017 with other platform support announcements in the future Complete hardware and software solution utilizing DRAM for application performance, a tier of flash for persistence and a backup power source to move data from DRAM to flash. Supports up to 1TB of capacity Up to 20x reduction in database restart times Up to 27x faster checkpoints 1 HPE Internal Labs test. HPE Scalable Persistent Memory, restarting 1000GB Hekaton Database is as fast as restarting 200GB database or 20x, November 2016 2 TPC-C Benchmark Throughput with Checkpoint (trans/sec). Calculated Time to Checkpoint and Restore a Docker Container running MySQL, compare Persistent memory vs. SSD, November 2016 Supported on HPE ProLiant DL380 Gen10 876401-B21 HPE Scalable Persistent Memory 256GB 876402-B21 HPE Scalable Persistent Memory 512GB 876403-B21 HPE Scalable Persistent Memory 1024GB Available as HPE Factory Configure to Order (CTO) SKUs 1 HPE Internal Labs test. HPE Scalable Persistent Memory, restarting 1000GB Hekaton Database is as fast as restarting 200GB database or 20x, March 31, 2017 2 TPC-C Benchmark Throughput with Checkpoint (trans/sec). Calculated Time to Checkpoint and Restore a Docker Container running MySQL, compare Persistent memory vs. SSD, March 31, 2017 Learn more: https://www.hpe.com/info/persistentmemory

Planned Platform Support HPE 16GB NVDIMM Right-sized persistent memory for smaller storage bottlenecks Ideal for: Smaller database storage bottlenecks, write caching tiers and any workload constrained by storage bottlenecks Key Features: DRAM-level performance for fastest performing persistent memory Up to 384GB total capacity in a single server HPE Smart Storage Battery for backup power 2x the capacity of 1st generation HPE NVDIMMs4 Up to 50% reduction in database licensing costs using the fastest tier of storage on HPE servers3 OS Support: Microsoft: WS2012 R2 (HPE driver) and WS2016 (inbox driver) Linux: RHEL 7.3 and SLES 12 SP2 HPE 16GB NVDIMMs are flash-backed DIMMs designed to eliminate smaller storage bottlenecks while delivering the performance of memory with the persistence of storage. HPE 16GB NVDIMMs are flash-backed DIMMs designed to eliminate smaller storage bottlenecks Ideal for smaller database storage bottlenecks write caching tiers and any workload constrained by storage bottlenecks. Up to 192GB in a 2s server (16GB x 12 NVDIMMs) and up to 384GB in a 4s server (16GB x 24 NVDIMMs) Double the capacity of 1st generation HPE NVDIMMs Software licensing reduction achieved through using NVDIMMs with fewer server core pairs (reduced core pair licensing from database vendors) vs. using block storage devices. Supported Platforms: DL360, DL380, DL560, DL580, BL460c, Synergy 480, and Synergy 660 Planned Platform Support ProLiant DL Gen10: DL360, DL380, DL560 and DL580 Synergy Gen10: SY480 and SY660 BladeSystem Gen10: BL460c 3 HPE Internal lab testing. Data gathered on pre-release hardware and software, final results may differ, Nov. 2016. 4 Comparing 8GB NVDIMM to 16GB NVDIMM equals 2X capacity increase, September 2017. Learn more: https://www.hpe.com/info/persistentmemory

HPE Scalable Persistent Memory Use Cases Storage at the speed of memory Large in-memory compute with data persistence Real Time Analytics Faster Checkpoints and Restores OLTP + OLAP = HTAP Use Cases for HPE Scalable Persistent Memory: TB-scale in-memory compute with persistence. Utilizes DRAM for application performance with an added level of persistence with a flash backup tier. Reducing database storage bottlenecks. Database performance can be vastly improved using low-latency DRAM to reduce or eliminate storage bottlenecks in lieu of using higher-latency block storage devices. Many databases have separate operations for the transactional portion of the database and the analytics portions which uses data tables from the transactional portion. Customers can utilize TB-scale HPE Scalable Persistent Memory to enable real time analytics – known as HTAP – where they do the transactions and analytics simultaneously. Software-defined storage implementations can also benefit from HPE Scalable Persistent Memory. From commercial applications such as FlashSoft or AutoCache to HPE StoreVirtual VSA and Microsoft Storage Spaces, HPE Scalable Persistent Memory can be used to eliminate storage bottlenecks in those implementations as well as be used as a fast caching tier. The concept of checkpointing is when an application is running in-memory and data in-flight needs to be periodically checkpointed to non-volatile media like HDDs or SSDs. This operation typically is a bottleneck for an application and can impact application performance. What if you did that checkpoint to HPE Scalable Persistent Memory – using DRAM on the nanosecond latency memory bus? This can dramatically improve overall application performance. We have tested this in HPE internal labs and have seen up to a 27x performance improvement. Restores using HPE Scalable Persistent Memory can also be a source of latency as you restore from HDDs or SSDs. When you restore from HPE Scalable Persistent Memory, you are restoring from DRAM with nanoseconds of latency on the memory bus. We have a documented use case where we restored a Microsoft SQL Server Hekaton in-memory database (200GB, running 500K Transactions Per Second (TPS)) using SSDs. That restore finished in a very respectable 20 mins. We then restored using HPE Scalable Persistent Memory and the restore went from 20 mins to 45 seconds! That’s an over 20x reduction in restore time! Anywhere you have storage bottlenecks today, where you are writing to block storage devices and using a layer of persistent memory in place you can see huge benefits. These are just a few of the many use cases for HPE Scalable Persistent Memory. Check www.hpe.com/info/persistentmemory often for the latest technical papers on emerging use cases. Reducing database storage bottlenecks Software-defined storage acceleration And much…much…more Big Data Service Providers Performance Tier Virtualization

Reducing Database Storage Bottlenecks Software Licensing Reduction HPE NVDIMM Use Cases Storage at the speed of memory Reducing Database Storage Bottlenecks Software Licensing Reduction Write Caching Up to 4x faster and better CPU Utilization with byte addressable storage. Up to 50% reduction in core-based software licensing with NVDIMMs Up to 2x faster with NVDIMMs than with NVMe SSDs Key Messages: HPE 8GB NVDIMMs support up to 16 NVDIMMs in a DL360 or DL380 Gen9 server featuring E5-2600v4 processors (Broadwell) for a total of 128GB of total capacity. NVDIMMs are ideal for eliminating small storage bottlenecks and are the fastest tier of storage on HPE ProLiant servers. Database Storage Bottlenecks – like transaction logs – are one of the biggest offenders of latency. That’s because the software is often waiting on the hardware to complete write operations. In SQL Server 2016, Microsoft changed the portion of the transaction log with the most recent writes – the Tail of the Log – to address HPE NVDIMMs as byte addressable storage. Now the transactions written to the log hit the DRAM on the NVDIMM greatly reducing the latency experience during transaction writes. This allows up to 4x faster performance and better CPU utilization (achieved through more efficient software operations from byte addressable software). 4x number compares writing transaction logs on SAS SSDs vs. NVDIMMs with SQL Server 2016 Tail of the Log byte addressable storage capability enabled. Software licensing reduction. We compared an HPE ProLiant server running with 32 cores and SAS HDDs compared to an HPE ProLiant server running with 16 cores and NVDIMMs. The 16 cores with NVDIMMs outperformed the 32 core system with SAS HDDs – resulting in a 50% reduction in server cores. Many software apps like Oracle DB license software by the server core pair. This can result in substantial savings on software licensing costs. NVDIMMs are also excellent at reducing latency and improving performance as a caching tier. By moving hot data writes to faster DRAM on the memory bus, applications that experience latency from writing hot data to slower media can benefit from NVDIMMs. Microsoft Storage Spaces was up to 2x faster with NVDIMMs as compared to using NVMe SSDs (80% reads/20% writes on 8GB NVDIMMs). HPE NVDIMMs are ideal for eliminating write storage bottlenecks and are the fastest tier of storage on HPE ProLiant servers. Storage Spaces