Over the past decade, the rapid evolution of next generation DNA sequencing technologies has inundated the field of Bioinformatics with sequence data. As these technologies have matured, and as more and more researchers have adopted them, Data repositories and researchers alike have become flooded with sequencing data. For most researchers, this presents a significant computational burden; the solution to which is not always obvious.
For those with occasional data analysis needs, the use of Cloud computing resources may address this burden. For many however, concerns about the data transfer rate when moving data to and from the cloud persist, as do issues regarding data management and security as well as cost if you are utilizing that resource 24x7. Those with a chronic computational burden often prefer a dedicated hardware solution, which might include a CPU cluster or some alternative ‘accelerator’ hardware. CPU-clusters are commonplace as they offer great flexibility but require significant administrative overhead for maintenance & upgrades. And all that flexibility makes for sub-optimal performance if you’re doing the same sorts of things over and over again.
Other dedicated ‘accelerator’ solutions include general-purpose Graphics Processing Unit (GPU) cards or Field Programmable Gate Array (FPGA) cards. GPU cards can be conveniently programmed using standard C programming language but they lack the performance FPGA cards provide. In particular, GPU cards offer very poor performance for Heuristic algorithms, which unfortunately, predominate in the field of Bioinformatics (think NCBI Blast or HMMER3.0). Furthermore, porting an algorithm to run on a GPU card requires significant tuning in order to achieve optimal performance. In other words, one cannot just take standard C-code and recompile it for GPU architecture and expect a significant boost in performance. Additionally, the Bioinformatics tools that are currently available for GPU cards are all open source and lack technical support, which can also be a burden for end-users. FPGA cards offer superior performance and power efficiency as compared to CPU-clusters and GPU cards. And since the power usage associated with a standard CPU cluster can be quite large when you factor in power & cooling costs, not to mention the cost of the space devoted to those resources, FPGA-based systems can accomplish huge monetary savings in power costs alone. And System Administrators love our hardware because it is comparably easy to maintain.
Most of our TimeLogic customers utilize both FPGA and standard CPU-cluster hardware in their compute resource facilities. In these cases, the ability to offload specific applications onto one of our DeCypher® servers frees up CPU-cluster resources, which can in turn extend their lives. And upgrading or expanding one of our DeCypher® solutions is a far simpler process than upgrading a CPU-cluster.
J-Series FPGA Hardware
Our J-Series Similarity Search Engine (SSE) was engineered to handle the explosion of data generated by next-generation sequencing platforms. With an eye towards Heuristic algorithms in particular, this new circuitboard design pushes the limits of the latest Xilinx® FPGA chips and the PCIe data bus. By augmenting available memory, not only have we been able to deliver huge improvements to Tera-BLAST performance, we can also provide an update to our popular DeCypherHMM algorithm that is based on the well-received HMMER3 package. And our new VelociMapper algorithm effectively runs at the limits at which the host server can read in a FASTQ data file and write out a BAM file. No other acclerators can come anywhere close to this level of performance. And you can find out for yourself by having us run custom benchmark tests using your own data.
In addition to hardware performance upgrades, our DeCypher software also includes job queueing and distribution, administrative tools, and application specific support from our team of experienced bioinformaticians. For additional details regarding DeCypher® host-server options, please see our product FAQs page.