Hewlett-Packard (HP) says it will address the I/O bottleneck created by internal and external storage in its next generation of ProLiant servers, boosting performance by up to 50% for popular applications such as online transaction processing (OLTP) and streaming video.
HP said in two separate blog posts that its upcoming ProLiant Gen8 server line will offer solid-state drives (SSDs) in combination with a data caching algorithm that it calls Dynamic Workload Acceleration.
The HP ProLiant servers include a 6Gbps embedded SAS controller that moves from using 15,000rpm SAS hard drives to SAS-based SSDs. The new controller offers six times the performance of previous generation controllers. DRAM cache capacity is also being doubled on the servers.
"Yes, Gen8 will be a screamer and workloads will run at remarkable levels," said Steve Schultz, who leads HP's Smart Storage Firmware Engineering team. "But beyond performance, storage capacity is another pain point."
ProLiant Gen8 will also support up to 50% more hard drives over the previous generation line, for up to 36TB of capacity in a single 2U (3.5-in high) DL380.
HPs strategy revolves around its Gen8 Smart Array RAID controller, which uses SSDs and the Dynamic Workload Acceleration to migrate the hottest data to the highest-performance storage.
HP claims that its Gen8 Smart Array RAID controller with SSDs is 60 times faster than the previous generation controller with hard disk drives. And, with PCIe Gen 3 support, bandwidth has been increased by more than 85%, the company said.
With its SSD option and Dynamic Workload Acceleration caching software, HP is making a play against other vendors who've also upped the ante with NAND flash and data caching, such as Oracle/Sun, Fusion-io and EMC, said Valdis Filks, Gartner's research director for information technology infrastructures.
"Storage really is one of the biggest bottlenecks. The server has increased in speed 10 times and disk storage has probably not doubled in the last five years as far as response times," Filks said. "I always say, this SSD market is a battle of the caching algorithms. It's about who has the best caching algorithms for pre-fetching data and being able to understand the data."
Fusion-io has been selling PCIe flash cards for servers for about four years. In February, EMC announced its VFCache product line, an initiative to sell PCIe-based NAND flash cards for servers as a caching element to increase I/O performance by up to 4,000 times. The company also began selling storage arrays filled with 10 or 15 of the flash cards and connected by InfiniBand networks for high-performance application workloads.
EMC's first server customer will be Cisco, which is planning to use the PCIe cards in its Unified Computing System Servers (UCS).
But SSDs in servers only solve part of the problem, according to Filks.
"At the moment if you stick this SSD card in the server, it really doesn't know what the storage array is doing. We believe within the next three years, they'll try to have software where the card communicates with EMC storage arrays, and when that happens EMC will have a lovely lock-in," Filks said, adding that HP will likely go down the same path.
But, EMC, HP and Fusion-io don't have a leg up on Oracle/Sun, which has been selling its Exadata Storage Server Software for databases for years and which communicates at the application level, Filks said. Because it is tightly integrated with the Oracle database, the software can pre-fetch data queries, giving it an advantage in terms better millisecond response times over storage systems, he said.
"Anyone at the storage layer is coming to the party late," Filks said. "Storage is so slow, that it's true people will get an advantage by using SSDs, but [EMC, HP and others are] really just trying to catch up."
In addition to supporting SSDs in its ProLiant server line, HP said it is offering something called SmartMemory and "flexible" 10GbE networking ports.
HP said that with SSDs, Dynamic Workload Accelteration, SmartMemory and 10GbE connectivity performance of some applications will increase by up to 50%, including transactional throughput for OLTP and video streaming. It also said the servers will use 88% less energy and 75% less floor space, and enable the deployment of 30% more virtual machines (VMs) with a reduction of 40% in energy-per-VM.
"We designed our memory to handle enterprise application workloads by increasing the bandwidth by 25% while offering 50% better capacities and 35% memory uptime improvement using our Advanced Memory Error Detection Technology," HP said in a blog. "Key goal for our memory design was to achieve power efficiency as we increased performance. All this translates to more virtual machines, database acceleration and efficient balanced system resource allocation to handle any workload."
Lucas Mearian covers storage, disaster recovery and business continuity, financial services infrastructure and health care IT for Computerworld. Follow Lucas on Twitter at @lucasmearian or subscribe to Lucas's RSS feed . His e-mail address is firstname.lastname@example.org.