IT/Datacenter & Super Computing - Page 11

Supercomputing 2014: In the world of HPC (High-Performance Computing) the bleeding edge is always the preferred route to realize insane computational power. HMC (Hybrid Memory Cubes) are the next big thing, and offer plenty of performance advantages over existing DRAM. The current generation of HMC technology sips power and provides more density and performance than existing memory technology. With 15 times the performance, 90 percent less space, and 70 percent less power consumption, it is easy to see why industry leaders are touting the advantages of HMC. The key to HMC adoption, as with any new technology, lies in the committees that establish industry-standard interface specifications.

The HMCC (Hybrid Memory Cube Consortium) was founded by Micron, Altera, Open-Silicon, Samsung and Xilinx in 2011 and has grown to more than 150 members. At Supercomputing 2014 the HMCC has announced the finalization and public availability of the HMCC 2.0 specification.

The new specification increases speed from 15Gb/s to 30Gb/s and migrates the associated channel model from short reach (SR) to very short reach (VSR). VSR is key to the eventual fusion of HMC into the CPU. The path to faster data processing, as with storage, involves getting closer to the CPU. The future integration into the CPU will expand upon the tremendous performance advantages of HMC. Imagine an L2 cache with 100 times the capacity.

Supercomputing 2014: Intel and SGI combined their talents to create an HPC monster that touts 30 million IOPS of 4k random speed with 180GB/s of sequential throughput. Scaling storage performance and capacity in tandem is an ongoing challenge in the enterprise storage world, and old interfaces have been the primary culprit hampering these objectives. A diminishing point of returns is reached as more storage devices are added to the server, and performance begins to decline as latency increases. This is a particular pain point when utilizing RAID and HBA architectures in tandem with 2.5" SSDs.

Enter the PCIe SSD. Moving flash to the PCIe bus provides better performance scaling, but many initial revisions of PCIe SSDs leveraged existing standards, such as AHCI, for host communication. This leads to performance degradation and excessive CPU overhead as performance scales. As explained in our Defining NVMe article, NVMe is a new storage protocol designed specifically for non-volatile memory. A slew of architectural refinements combine to provide the best performance possible over the PCIe interface. Intel's DC P3700 (covered in-depth in our Intel DC P3700 1.6TB NVMe Enterprise Review) is one of the fastest PCIe SSDs available, and the combination of NVMe and consistent performance provide enhanced scalability when deploying multiple units.

Intel and SGI decided to push the performance envelope by integrating a whopping 64 of the DC P3700's into a 32-socket server to test the limits of NVME scaling. The results are quite impressive, and the modified SGI UV 300H SAP HANA server surpassed 30 million IOPS in 4k random testing. Perhaps the most impressive aspect is the linear performance scaling in the graphic above. The solid blue lines denote the IOPS performance curve as more P3700's were added, and the dotted blue line is a trend line illustrating a theoretical linear progression. From the results we can see that the SGI beast barely deviates from the perfect scaling trend line. The throughput, tested with 128k sequential blocks, topped out just shy of 180 GB/s.

Supercomputing 2014: Intel has announced a new Xeon Phi processor, code-named Knights Hill, at Supercomputing 2014. The Xeon Phi co-processors are the offspring of the Larabee project, and Intel has improved performance and inter-operability with each successive product generation. Knights Hill is a key advancement for Intel in the HPC (High-Performance Computing) market, and will leverage a 10nm process and integrate Intel's Omni-Path Fabric technology.

The Knights product series competes directly with NVIDIA in the supercomputing market, and Knights Hill is a natural progression of the product line which will enhance performance scaling and bandwidth while simultaneously reducing power consumption.

Knights Landing features the Intel Silvermont Architecture that is designed specifically for HPC applications. The architecture significantly boosts single thread performance by 3X in comparison to the Knights Corner product. The on-package memory tops out at 16GB and offers amazing bandwidth, over 5x more than DDR4. This stacked memory design also significantly reduces the power consumption of the memory subsystem by as much as 5X.

Supercomputing 2014: Huawei announced their new FusionServer X6800 products at Supercomputing 2014. The new FusionServer X6800 supports a variety of high-density nodes in a 4U chassis, which provides flexibility for different configurations of compute, storage, and GPU acceleration within the same server. The X6800 also supports the latest SSD configurations and offers two to four GE ports or two 10Gbe ports.

Huawei also announced a new liquid-cooled version of the FusionServer E9000. This blade server features cutting-edge heat dissipation technology in a 12U enclosure. This liquid-cooled behemoth features 16 slots and redundant power supplies, along with 40GBe or InfiniBand FDR (56G). The platform is built upon a scalable architecture that allows for dual or quad CPU nodes. Huawei will also be demonstrating the RH5885H V3 server, a standard 4U rack server that supports four Intel E7-4800 v2 processors, and the RH2288 V3, a dual socket 2U server rack that supports Intel E5-2600 v3 processors.

Foremay, Inc. has announced their new 4TB and 8TB SATA 6Gb/s SSDs. SATA SSDs are one of the hottest growth segments in the datacenter, and more capacity is always welcome. There are SAS SSDs with up to 4TB of capacity, such as the SanDisk Optimus MAX, but the introdcution of a high-capacity SATA SSD will broaden the use-case for high-capacity SSDs. Foremay focuses on the OEM market, and the new SSDs are geared for servers and ruggedized applications. The EC188 and SC199 features a random read/write speed of 100K/100K, and sequential read/write speeds of 500/500 MB/s.

The SSDs feature AES-256 encryption and are TCG Opal 2.0 compliant and feature military-grade secure erase functionality. The new SSDs are designed for high heat environments to provide reliability in enterprise and industrial applications. Foremay SSDs are custom built, and can also equipped with a supercapacitor to provide power loss protection. There simply aren't any SSD controllers on the market to address that much flash, so this is likely to be a custom solution with a bridge device akin to the Intelliprop Hydra SATA bridge we evaluated recently.

OCZ Storage Solutions has just announced the release of their Vertex 460A. The original Vertex series has been a stellar product with a history that spans back to the original version with the first-gen Indilinx Barefoot controller. The new version leverages Toshiba's latest A19 MLC NAND flash. The A19nm process geometry is the second generation of Toshiba 19nm MLC. The new version also features the Barefoot 3 controller and sequential speeds of 545/525 MB/s read/write (480GB model). Random speeds also top out at 95,000/90,000 random read/write IOPS, respectively, but performance varies depending upon capacity, as noted in the graphic below.

The Vertex 460A features an endurance rating of 20GB of writes for the three-year warranty period. OCZ is providing their new ShieldPlus warranty, which provides advance shipping and covers return shipping costs if there is the need for an RMA. The new Vertex also features Acronis True Image for cloning an existing installation to the SSD, and a 3.5" desktop adaptor. OCZ recently launched a new online shop, and we expect units to be available there shortly.

The new Vertex comes in 120, 240, and 480GB capacities and also offers toll-free tech support. Chris put the original Vertex 460 through the testing gauntlet, and the SSD received the TweakTown Editor's Choice Award. The switch to the newer process shouldn't yield huge variations in performance, as a matter of fact the specifications are very similar. Head over to the OCZ Vertex 460 240GB SSD Review for a closer look at the previous model, which should be indicative of performance experienced with the new model.

Supermicro has announced world-record setting performance in the STAC-N1 and STAC-A2 benchmarks. Supermicro's 3rd generation Hyper-Speed platform, in tandem with the Intel Xeon Phi co-processor, delivered the record-setting results from an incredibly slim 1U server. The results indicate a propensity for superior performance in HFT (high Frequency Trading) applications, which crave low latency and high performance. Performance consistency is a big key to delivering predictable and sustainable QoS for HFT applications. The Hyper-Speed platform nailed key requirements with the lowest mean latency, max latency, and jitter. The slim 1U platforms can be deployed with up to 3 Intel Xeon Phi co-processors, expanding the use-case for even more demanding workloads.

The Hyper-Speed Ultra provides an impressive stable of connectivity with 10 2.5" drive bays, 8 12Gb/s SAS 3 ports, 2 SATA 6Gb/s ports, and a range of PCIe connections. The platform also accepts 2 NVMe drives via the AOC-URN2-i2XT. The system is powered by dual E5-2643 v3 Haswell processors. Supermicro has several high-performance systems available, and the pending release of their ULLtraDIMM-enabled platforms may increase performance beyond their own world record. 3

We recently had a chance to take a Supermicro development system for a spin in our SanDisk ULLtraDIMM DDR3 400GB SSD Enterprise Review. We also feature full evaluations of a number of Supermicro and competing platforms in the Motherboard and Server categories of our IT/Datacenter section.

Power consumption is the highest ongoing expense in the datacenter, and for giants like Facebook it can easily add up to billions of dollars per year. One of the most obvious sources of power consumption spawns from cooling costs. Power consumption generates heat, and Facebook has grabbed the low-hanging fruit by moving to open-air datacenter designs that radically reduce cooling requirements. Now Facebook has turned their attention to UPS systems for the next layer of power savings. Reducing overall power consumption is key because it also incurs the expense of power backup. During a power loss event the systems automatically fall back to massive UPS systems that provide enough power, typically 90 seconds worth, to cover the gap until backup generators come online. Facebook has already altered UPS design by migrating from large central UPS systems to seven-foot tall server cabinets interspersed throughout the datacenter.

Today these massive power backup systems rely upon lead-acid batteries, but now Facebook is experimenting with the same type of lithium-ion batteries found in today's latest electric vehicles. The long term cost of maintenance is lower for lithium-ion batteries, and they also deliver more power in a smaller footprint. Facebook is experimenting with designs that embed lithium-on batteries at the rack level. Two batteries will slide into each rack and provide UPS protection. This design also reduces the chance of UPS failure. If a standard centralized UPS fails the entire datacenter can go down. With rack-level battery backups, only small groups of servers would be effected by individual failures.

Recent advances in lithium-ion battery technology have been fueled by electric car development. Vehicles like the Volt, Tesla, and Leaf, have ushered in advanced battery technology and also lowered the overall cost. Now that cost of Li-on batteries has fallen they have become a sensible alternative for UPS applications in massive datacenters. Facebook is integrating their new designs into their Open Compute initiative, which might serve to expand the widespread use of Lithium-ion in the datacenter. There is no word on how increased demand would affect the overall pricing.

Open Server Summit 2014 focuses on next-generation server designs that leverage industry-standard hardware and open-source software. The show is a great place to view future server technology, which makes it the perfect venue for displaying the Diablo MCS (Memory Channel Storage) architecture at work on the SanDisk ULLtraDIMM. The SanDisk ULLtraDIMM DDR3 SSD brings latency as low as five microseconds by sidestepping the traditional storage stack, and communicating via the DDR3 bus. This reduces cabling, complexity, and components required for typical storage deployments.

The slim form factor, which takes advantage of the existing memory subsystem, will enable radical new server designs, particularly in the blade and microserver segment. The hardware consists of a JEDEC-compliant ULLtraDIMM that presents itself as a block storage device with 200 or 400GB of capacity. The ULLtraDIMM utilizes two Marvell 88SS9187 controllers running the Guardian Technology Platform to increase endurance and reliability. This tandem delivers random read/write performance of 140,000/40,000 IOPS, and sequential read/write speeds up to 880/600 MB/s. Ten DWPD (Drive Writes Per Day) of endurance, and a five-year warranty (or TBW) are provided by SanDisk 19nm eMLC NAND.

The real genius of the ULLtraDIMM design is its enhanced parallelism. Stacking several devices in parallel unlocks key performance advantages that will challenge even the fastest datacenter-class PCIe SSDs. We recently had a chance to take an in-depth look at the ULLtraDIMM and post our independent third party testing results in the SanDisk ULLtraDIMM DDR3 400GB SSD Enterprise Review. Head over to the PCIe category in our IT/Datacenter section for a look at competing PCIe devices.

Lately HDDs aren't gaining in capacity as quickly due to the limitations of PMR (Perpendicular Magnetic Recording). PMR stores magnetic bits of data vertically, allowing manufacturers to cram more data onto the HDD's platters, which provides more density than the previous horizontal method. Every new technology has its limits, and PMR has nearly reached the end of its evolutionary cycle. Now manufacturers are turning to HAMR (Heat-Assisted Magnetic Recording) to increase density. HAMR uses a small laser to heat the surface of the platter to 800 degrees Fahrenheit before data is written. The laser is incredibly small and embedded into the drive's write head, and the small heated surface area cools back down in under a nanosecond.

Heat alters the magnetic properties of the disk for this nanosecond in time, and removes or reduces the superparamagnetic effect while data is written. This process allows for exponential gains in density, and HAMR drives with up to 20TB of storage are on the horizon. While this technology sounds a bit far-fetched, working development drives have already been displayed. With any new technology one of the immediate concerns is a lack of development tools. A team from A*STAR, led by Hongzhi Yang and the National University of Singapore, have designed a pump-probe laser to test HAMR devices. This allows accurate testing of temperature-dependent recording in localized regions without actually destroying the media. This is one more step on the path to creating affordable HAMR HDDs, and the first Seagate HAMR HDDs are projected to release in 2016 timeframe.