Home 
SERVICES GNU/Linux
SERVICES GNU/Linux
| GNU/Linux |
|
The Gluster Distribution The Mazda Technologies GNU/Linux Engineering team with its years of experience working on large cluster projects is developing a unique cluster distribution called "Gluster". Gluster is virtually platform and architecture independent and can be used to build SMP-like machines on clustered architecture, and also to build Clustered File Storage that scales to peta bytes of storage. The team has worked to replace the traditional “Big-Iron” machines with reliable and affordable compute clusters built on COTS, Free Software and Open Standards. This gives the end user a solution comparable in performance and reliability to the Big-Iron SMP machines but at a fraction of the cost. Gluster is a cluster distribution that helps seamlessly build scientific computing clusters, NUMA machines out of Clusters, Super-storage clusters, Database clusters and Enterprise System Imaging. Gluster NUMA: As of today, NUMA type systems are very expensive and proprietary. Their cost increases exponentially along with their size. While NUMA vendors are currently tackling the problem of moving to commodity Itanium2 and Opteron processors to cut down the cost, Our team takes a radically different approach to resolve the multi-processor scalability problem by working with the VM & Scheduler in the Linux Kernel, it is a software approach that does not re-write of the applications. Clustering backbone has evolved from simple 100Mbps PCI Fast Ethernet to double-data-rate 20Gbps PCI-E Infiniband HBAs with 2.x latency. This improvement helps clusters yield NUMA class efficiency. “Thunder” supercomputer with 4096 processors demonstrated 86.7% efficiency which was higher than the comparable NUMA type supercomputers. Gluster adds system-wide process name space, shared memory, process/thread scheduler, high performance streams migration, intelligent headless platform manageability and automated system image provisioning features to this supercomputing class hardware. For example, Gluster SSI cluster of Intel’s TS4850HW4/M (512 cores based on Paxville-MP Xeon 3.3GHz, independent 800MHz FSB per 2 processors, 2TB DDR2 RAM, 64 Ultra320 2x120GB mirrored scratch disk storage and 64 20Gbps DDR Infiniband PCI-E HCAs, IPMI 2.0 platform manageability) systems will provide a mini supercomputing class performance to meet most scientific and high-end enterprise business demands at a fraction of the cost of the comparable SSI/NUMA machines. Gluster FS: Centralized SAN based storage is increasingly becoming a bottleneck for many data centers. Tens and Hundreds of servers simultaneously demanding data over a single channel easily chokes the bandwidth. Lot of compute cycles were wasted waiting for I/O. We have solved this problem while building supercomputers by implementing clustered file storage. This innovation is now scaled down to meet data center requirements too. Mazda Technologies clustered storage (GlusterFS) can scale to peta bytes of data with linear scalability and reliability. For example, Thunder supercomputer was backed by 200TB of storage with a 16 node cluster at 6.4GB/s parallel I/O performance. Current product lines include Infiniband based backbone and hot-swap SATA II disks which can accommodate huge amount of storage and I/O bandwidth in a relatively smaller form factor. Linux Kernal Mazda Technologies engineering group can provide Linux kernel modifications, consulting, and development support. Our team includes talented kernel developers, so you can tap their collective experience to solve your Linux kernel porting/stability/feature issues. Mazda Technologies kernel team has experience with both 32-bit x86 and 64-bit Itanium 2 architectures and has also done embedded development with the Strong ARM processor. Kernel-level professional services work usually arises from a desire to resolve customer issues with stability and feature addition. This can includes back porting needed functionality from modern kernels into ISV-supported kernels and distributions, as well as deploying features from experimental kernels into production environments to, for example, enable needed functionality during a product's development phase. Embedded kernel work usually supports specific products using Linux and often includes development of product-specific drivers, as noted below. Driver Development Mazda Technologies engineers can develop Linux device drivers for various I/O, video, RAID, and specialized cards and systems. These engagements typical arise from customers wanting to add Linux functionality to an existing product or who are developing a consumer device with specific hardware than runs embedded Linux. Performance Optimization Mazda Technologies software team can identify and resolve code or configuration bottlenecks ranging from kernel configuration to system configuration to site architecture to database structure. These engagements typically result from customer realization that something is amiss with application performance, scalability, or sustainability. For example, analysis may indicate that system RAM is being completely used, or stability decreases when SMP kernels are employed or certain database operations do not scale effectively. Custom Software/Applications Development Mazda Technologies Professional Services has world class capabilities for development of custom open source software. This capability ranges from building core Linux components to a wide range of applications including embedded technologies, GNU-based tools, web server and e-commerce, content and collaboration management, and web application development. The Mazda Technologies Linux Engineering group has worked with numerous Tier 1 customers in delivering their custom software needs. Mazda Technologies Linux Engineering group will work with you to understand your specific needs and then deliver upon the commitments given. Applications Porting a Linux-based infrastructure. By outsourcing this development, your team can remain focused on its feature and product roadmap while allowing Mazda Technologies highly-skilled device team to enable new platforms for your products. Mazda Technologies specializes in 64-bit Linux computing as well as application parallelization, i.e. the re-writing of applications to execute efficiently in SMP or other distributed-processor environments. (HPC) High Performance Computing Mazda Technologies has bioinformatics and computational physicists that understand your applications. They can help you parallelize your code, optimize applications for massively-parallel execution on Beowulf clusters, remove bottlenecks by understanding the computational complexity of your applications, and provide sophisticated consulting and development to help you unlock the power of your Linux-based computational infrastructure. Engineering Strengths Mazda technologies Linux Engineering team includes specialists from various fields.
|