Code Sport


Over the next few columns, we will take a look at data storage systems, and how they are evolving to cater to the data-centric computing world.

Last month, we featured a special edition of CodeSport, which discussed the evolution of programming languages over the past 10 years, and how they are likely to evolve over the coming 10 years. The article went on to hazard a guess that the ‘Big Data’ explosion would shift the momentum to languages that make data processing simple and efficient, and make programs data-centric instead of the code-centric perspective. Many of our readers had responded with their own views on how they see computing paradigms evolving over the coming 10 years. Thanks a lot to all our readers for their feedback and thoughts.
One of our readers, Ravikrishnan, sent me a pertinent comment, which I want to share: “Thank you for your article on the evolution of programming languages. Indeed, there is a heavy momentum towards processing huge amounts of data using commodity hardware and software. While the basic concepts and algorithms of computer science would continue to hold sway, the sheer scale of the data explosion would require programmers to understand and apply algorithms where data does not fit in main memory. Hence programmers need to start worrying about data latency of secondary storage such as flash SSD/disk storage systems. In a way, the shift towards data-centric computing means more intelligent storage systems, and a need for programmers to understand about state-of-the-art storage systems, where big data is stored, processed and preserved. While this is not a traditional topic covered in CodeSport, given the importance of data storage in a Big Data world, it would be great if CodeSport does a deep-dive into state-of-the-art storage systems in a future column.”

It was a timely reminder for me. While I have discussed various Big Data computing paradigms in some of our past columns, I have not covered storage systems at all. So over the next few columns, I am going to discuss storage systems, and how they have evolved over years to cater to the Big Data explosion. I will take readers through some of the challenging problems as well as the state-of-the-art research directions in this space.

A storage systems primer
Let us start our journey into storage by understanding some of the basic concepts and terminology. In the traditional view of storage, we all know about the triumvirate of the CPU, memory and disk, where the hard disk (also known as secondary storage) is part of, or directly attached to your computer, and acts as the permanent storage. From now onwards, when I use the term storage, I actually imply the traditional secondary storage, which acts as the backup to the main memory (which is the primary storage). These include hard disk drives, flash/SSD storage, tape drives, etc.

Traditional HDDs are accessed using a variety of protocols such as SCSI, ATA, SATA and SAS. SCSI stands for Small Computer System Interface and is a parallel peripheral interface standard widely used in personal computers for attaching printers and hard-disks. ATA is another interface used for attaching disks; also known as IDE, wherein the controller is integrated into the disk drive itself, ATA is also a parallel interface like SCSI and both have their equivalent serial interfaces namely, Serial SCSI (abbreviated as SAS) and Serial ATA (abbreviated as SATA) which allow a serial stream of data to be transmitted between the PC and the disk drive.
Note, however, that in the traditional view of storage, it is part of the compute server, since it is directly attached to the server and is accessed through it. It is not an independent addressable entity, and is not shared across multiple computers. Typically, this is known as Direct Attached Storage (DAS). Access to the data in secondary storage is through the server to which it is attached; hence, if the server is down due to some failure, the data becomes inaccessible. Also, as data storage requirements increase, we need to have greater storage capacity. We produce 2.5 quintillion bytes of information every day, out of the Web searches we do, the online purchases we make, the mobile calls we make, and the social network presence we have (a quintillion is 1000 x 1000 x 1000 times a billion).

Given the volume of Big Data that gets produced, the storage requirements go on increasing exponentially. However, in case of Direct Attached Storage, the number of I/O cards (for example, SCSI cards) that can be connected to a computer is limited. Also, the maximum length of a SCSI cable is 25 M. Given these restrictions, the amount of storage that can be realised using the conventional directly attached storage is limited. Also note that DAS results in uneven storage utilisation. If one of the servers has used up all its disk storage and needs further storage, it cannot use any free storage available in the other servers.

As opposed to the traditional server-centric paradigm we have seen above, in a storage-centric view, storage exists as an independent entity, apart from the compute servers. Storage can be addressed independently, from multiple servers.  Though the storage is an independent entity on a network, to the operating system running on the compute server, it appears as if locally attached to the compute server.
Two popular forms of storage-centric architectures are Network Attached Storage (NAS) and Storage Area Network (SAN). The latter (SAN) allows storage entities to exist in a network, which can be accessed from compute servers using either special protocols such as Fibre-Channel or standard TCP/IP protocols such as iSCSI (internet SCSI). SANs provide block-level access to storage, just like traditional locally attached storage. In contrast, in the case of Network Attached Storage or NAS, a dedicated storage computer exists as an entity on the network, and is accessible from multiple compute servers concurrently. Unlike a SAN, a NAS provides file-level storage semantics to multiple compute servers, appearing as a file server to the operating system running on the compute server. Internally, the NAS file server would access the physical storage at block level to access the actual data, while this is transparent to the OS on the compute server, which is exposed only to file-level operations on the NAS server.
Hybrids of SAN and NAS also exist. Since there is no file system concept for SANs, various file protection and access-control mechanisms need to be taken care of in the OS running on the compute server. In case of NAS, file protections and access control can be enforced at the NAS server. The next concept we need to understand in the storage domain is Scale-up Storage vs Scale-out Storage, which we will discuss next month.

Remembering Aaron Swartz
It has been almost two months since the death of Aaron Swartz. Most of us would have read about the enormous outpouring of grief caused by this tragic loss. Aaron Swartz was a programmer first and foremost, and the reason I wanted to mention him in our column was not just because he was a well-known activist who fought for the freedom of information on the Internet, but because he is a sterling example of what differentiates a great programmer from the run of the mill. He had an enormous enthusiasm for building software that solves challenging problems. He was involved in the development of the RSS format, wrote the framework, and was a technical architect of, just to mention a few examples of his work. He had a great passion for expanding and sharing his knowledge with all developers. Rest in peace, Aaron.

My must-read book for this month
This month’s must-read book suggestion comes from one of our readers, Aruna Rajan. She recommends the book Introduction to Information Retrieval by Christopher D. Manning, Prabhakar Raghavan and Heinrich Schütze. This book focuses on various information retrieval techniques, including the most popular one of Web search engines. The book is available online at Thank you, Aruna, for your suggestion.

If you have a favourite programming book/article that you think is a must-read for every programmer, please do send me a note with the book’s name, and a short write-up on why you think it is useful, so I can mention it in the column. This would help many readers who want to improve their coding skills.

If you have any favourite programming puzzles that you would like to discuss on this forum, please send them to me, along with your solutions and feedback, at sandyasm_AT_yahoo_DOT_com. Till we meet again next month, happy programming and here’s wishing you the very best!


Please enter your comment!
Please enter your name here