February 04, 2008
It’s nearly impossible to overstate the importance of low-latency solutions in financial datacenters -- whether for risk analysis, electronic trade processing, or anything in between -- which explains GemStone Systems’ recent partnerships aimed at addressing the issue on two different fronts.
Made public in November, the first of these partnerships is a go-to-market arrangement with Azul Systems aimed at addressing the multiple needs of users who need to improve latency, scalability and performance in the face of ever-increasing data volumes. By approaching potential customers together and touting the benefits of running GemStone’s GemFire Enterprise Data Fabric on top of the Azul Compute Appliance, both companies expect to see increased interest in their solutions. GemFire lets users create the types of highly available in-memory data grids necessary for extreme transaction processing (XTP) environments and Azul’s Compute Appliances supply the memory -- ranging from 48 to 768 cores and from 96GB to 768GB of memory.
According to Asad Ali, senior product marketing manager at GemStone, testing between the two companies has shown that uploading an existing Java application into the GemStone-Azul environment is simply a matter of changing a few parameters, but the results belie the ease of deployment. Hundreds of commodity machines, he said, can be consolidated into several Azul appliances, and the combination of the proximity of the data and application, the reduction in networking, and the data being in-memory drastically speeds data access. Additionally, noted Ali, Azul’s no-pause garbage collection means Java applications can experience performance and scale the likes of which they rarely see in other environments.
Sanjay Sadhu, vice president of business development at Azul, is equally optimistic about the partnership, saying that a joint GemStone-Azul platform is actually greater than the sum of its parts. The financial services industry, he said, has serious latency concerns -- on top of scalability, simplicity and availability concerns -- yet many customers, including some top banks, “are not aware of a solution to the pain.” Combining Azul’s performance benefits with GemFire’s data access benefits not only solves these problems, but actually opens up new markets for institutions in terms of the number of traders who can access the data, said Sadhu. Azul Appliances can handle the needs of multiple applications because of the systems’ built-in SLA management, and GemFire allows multiple applications to access the data simultaneously.
Sadhu said the GemStone and Azul have been seeing a lot of interest in the joint solution, with most of it (perhaps not surprisingly) coming from the financial sector and focused on the data caching aspect -- a critical element of the growing XTP market. However, both Sadhu and Ali agree that other vertical markets definitely could stand to benefit. Sadhu cites telco as one potential market, while Ali notes they are seeing growth in e-commerce and online gaming, as well. XTP is extremely important in transaction-intensive Web applications, a point Ali said was driven home by the number of retail Web sites that went down over Thanksgiving weekend.
Aside from handling the requirements of extreme transactional environments, Sadhu said a GemStone-Azul platform also helps give shape to grid and distributed environments by putting the focus on the application instead of on managing the environment. He can’t say whether the software or the hardware leads the solution, but together, he summarized, the companies solve problems -- from giving power to Java applications to minimizing datacenter cost and complexity -- that appeal to all levels of decision-makers. “We do see our value prop to be in the eyes and ears of CIOs and the architects who are building the solutions,” said Sadhu.
More recently, GemStone announced a partnership with IBM and Intel (and, to a lesser degree, Cisco) meant to address the extreme low-latency needs of the electronic trading industry. Called GemCache, the joint solution consists of GemStone’s GemFire software, IBM’s BladeCenter HS21 and Intel Quad-Core Xeon processors. In multi-server environments, the solution has utilized Cisco InfiniBand switches. According to Gideon Low, principal architect for business development and alliances at GemStone, GemCache is a result of common customers asking the three vendors to create a tight solution that addresses the customers’ real-world needs.
In order to demonstrate GemCache’s real-world throughput and latency capabilities, the companies put it through a first-of-its-kind benchmark test conducted by the Securities Technology Analysis Center (STAC). The solution demonstrated zero-loss reliability and guaranteed messaging into the transaction. As far as numbers go, the official announcement of the benchmark testing states:
The benchmark results showed the best-of-breed solution was able to support latency of less than 1 millisecond (987 msec) at 6,000 orders per second, with full data replication using a two-server GemCache configuration. In the tests, GemCache was also able to deliver a mean latency of 519 microseconds at 6,000 orders per second with a single server without replication. Additionally, three active/active two-server clusters supported 18,000 orders per second with approximately 1 millisecond (1,017 msec) of end-to-end latency using six IBM blade servers.
Near- or sub-millisecond latency times are needed to stay competitive in the trading markets, says Low, “but at the same time, you’re expected to be rock-solid, bulletproof, and never fail as a system and never lose data.” The numbers clearly illustrate that GemCache meets the former requirement, but it is the latter that often gets ignored. With GemCache, however, Low says data loss is all but eliminated via synchronous copying, and he said that they actually tightened the constraints in the benchmark test beyond what most customers do in their datacenters. Even pulling the plug on the primary processing machine did not result in data loss or impact on external applications, save for minimal failover time, he explained. Previous generations of solutions generally are several milliseconds slower, added Low, and cannot combine scalability, throughput and low latency like GemCache, nor can they generally handle both the database and messaging tiers like GemCache does.
Peter Lankford, founder and director of STAC, said the kinds of latencies demonstrated by GemCache are “significant” to end-users with whom his firm has spoken -- users who usually are looking at roundtrip latency times in the low tens of milliseconds but want to go lower. Even more significant, said Lankford, is that GemStone, IBM and Intel did not bypass functionality in order to achieve an impressive number, but instead tried to construct a realistic benchmark with no data loss. Although this is the first execution-related benchmark STAC has published, Lankford added that it has created a standards organization over the past few months “to get everyone on the same page,” and the GemCache benchmark is a step in the right direction.
As for what drives what in a solution like this, Low is more definite than his more diplomatic GemStone colleague Sadhu. While Low acknowledges that hardware advances have been tremendous, he firmly believes the biggest difference maker in solutions like GemCache is the software architecture. Trying to carry out electronic trading transactions would be drastically slower using a traditional disk-based database architecture, he explained, so the move to an in-memory environment is critical -- although having fast hardware certainly doesn’t hurt.
Like the GemStone-Azul solution, GemCache is available and in production today, although the pieces must be purchased individually. However, said Low, the companies involved are working on plans to co-market and co-sell the complete solution. As for the differences between the two environments, Low, who had a big part in the early testing work with GemFire and the Azul appliances, explained that the major differentiators are latency times and the amount of data that can be managed. While a GemStone-Azul environment cannot match the latency times of GemCache, the former, is very good for caching and managing a couple of hundred gigabytes of data while maintaining a better-than-average latency rate. As a result, said Low, GemCache is ideal for carrying out electronic trade processing, whereas the GemStone-Azul platform is better for real-time risk management or other processes that require the ability to drill into large amounts of data very quickly.
Regardless of what a customer’s needs are, though, “[low latency] gets more and more important every day,” said STAC’s Lankford, “and the reason is that in this market, invention is the mother of necessity. As soon as somebody down the street deploys some new technology that decreases their latency, everybody else has to keep up or they start losing deals.” Any delinquencies in latency are only exacerbated by the ever-increasing volumes of data and transactions firms are facing, as well as by the reality that machines are doing so much more trading than people, he added. “Once you make the decisions in silicon instead of neurons, now you’re on a completely different timescale.”
10/30/2013 | Cray, DDN, Mellanox, NetApp, ScaleMP, Supermicro, Xyratex | Creating data is easy… the challenge is getting it to the right place to make use of it. This paper discusses fresh solutions that can directly increase I/O efficiency, and the applications of these solutions to current, and new technology infrastructures.
10/01/2013 | IBM | A new trend is developing in the HPC space that is also affecting enterprise computing productivity with the arrival of “ultra-dense” hyper-scale servers.
Ken Claffey, SVP and General Manager at Xyratex, presents ClusterStor at the Vendor Showdown at ISC13 in Leipzig, Germany.
Join HPCwire Editor Nicole Hemsoth and Dr. David Bader from Georgia Tech as they take center stage on opening night at Atlanta's first Big Data Kick Off Week, filmed in front of a live audience. Nicole and David look at the evolution of HPC, today's big data challenges, discuss real world solutions, and reveal their predictions. Exactly what does the future holds for HPC?