Talk:Fixed-block architecture

CKD versus CHR/CTR

The article mistakes terminology:

IBM mainframe systems' disk layout had traditionally been count-key-data (CKD)…

In fact, the physical layout of the first FBA devices remained precisely the same; it was in the disc controllers that the differences occurred.

CKD had implications for the channel programmer, in that it was possible for a record to be located not solely by its location, but by a key. However, the actual architecture was CHR (cylinder-head-record), also called CTR (cyl-trk-rec). See DASD article.

--UnicornTapestry (talk) 01:06, 7 January 2009 (UTC)

What is FBA

the article was a mess and I have tried to start to fix it. Is there any substantive difference between FBA and LBA

The RAMAC used fixed sector sizes but I am not sure how the file system accessed it nor how the storage stack was layered. Most of every other early computer that I know of used fixed sector sizes and again i am ignorant of their storage stacks. IBM abandoned fixed sector sizes in most mainframe models with the S/360 beginning 1964. The notable S/360 M20 used a fixed sector size but addressed them with the standard IBM CHR channel commands so it was not an FBA system or drive unless FBA means fixed sector sizes.

To the best of my recollection Fixed Block Architecture is a term invented by IBM with the introduction of the IBM 3310 and a small system (s/34??) sometime in the mid-70s to describe a fixed sector sizes in its HDDs. At that time and into the 1990s IBM used CKD, a variable record length architecture on virtually all of its mainframe HDDs. The rest of the world and I think IBM's small systems used fixed sector sizes.

I know of no other usage of "Fixed Block Architecture" prior to the 3310. To the best of my recollection and search, IBMs other products that used fixed sector (or block) sizes called them just that and did not use the term FBA. I may be wrong and would appreciate other editors input on the origins of this specific term as applied to storage.

The other aspect of FBA is the addressing as a linear contiguous set of blocks. This really isn't a drive issue since drives today still have cylinders and heads and now a variable number of blocks per zone. With fixed blocks, someplace in the storage stack a file name has to be translated into a list or lists of blocks. I really don't know enough about file systems of any of the mainframe any mini computers to know whether the systems went thru a layer where the storage medium was viewed as a linear contiguous set of blocks to be resolved to CHS at a lower level. So it maybe that IBM was first in the 70s, but I doubt it. Again input from other editors would be appreciated.

I really think this is a controller issue, when what was the first controller that presented the device as a contiguous series of fixed blocks. I know this wss a part of the ANSI proposal in the late 70s and was a key feature of SASI. So the earliest FBA/LBA drive that I know of was likely the embedded ANSI drives or maybe the embedded SASI drives.

Comments Tom94022 (talk) 05:41, 9 August 2014 (UTC)

I do not agree that the difference between a count-key-data disk drive and an FBA disk drive is only a controller issue. The IBM 2311, for example, had different models for count-key-data and FBA controllers. I have put a reference to the relavent IBM manual under “Further reading” in the article—it provides a detailed description of the differences between the models.

I believe the term FBA is an IBM-ism, similar to DASD. IBM's definition, however, applies to some of their products predating the IBM 3310, so it seems appropriate to include them in this article.

The idea to number blocks linearly was a later development. Early devices were addressed using cylinder-head-record, so counting records in sequence required a conversion. I am not familiar with the internals of the IBM file accessing software, but DEC software regarded each storage device as a collection of blocks numbered from 0 to the number of blocks on the device minus 1. The file system software sent this block number to the device driver, which did the necessary divisions to convert it to cylinder-head-record. John Sauter (talk) 04:07, 10 August 2014 (UTC)

Thanks for the response. I never meant to say it is only a controller issue but I do think the controller is a necessary part of the story. Note the following sentence in the lede:

FBA devices do not use the traditional CHR addressing, but reference fixed-length blocks by number, much like sectors in mini-computers.

By this component of the definition the 2311-11/12 are not FBA devices.

If linear block addressing is a necessary component of FBA then doesn't the controller have to be a part of the device? The 2311-1 and the 2311-11 address the blocks in the same manner, CHR at the controller level. Most if not all early drive didn't know about records or blocks or sectors at all; the data are bit serial with records/sectors defined by address marks, sync bytes and relative bit counting. It seems to me, and I appreciate the dialog, that if FBA=LBA, then the drive must have fixed sectors addressed linearly and until maybe DEC Massbus, ANSI or SASI all drives had a primitive interface that knew nothing of records/blocks/sectors. On the other hand if FBA means only fixed sector size, then its not synonymous with LBA and the article needs to be corrected. Also that's pretty much how the whole world operated except for IBM mainframe DASD from 1964 until some time in the 1990s. Tom94022 (talk) 06:53, 11 August 2014 (UTC)

FBA is not the same as LBA—FBA still used three numbers to address a data block, called cylinder-head-sector. I have updated the article to make this clear, and added some structure to it.

Even though FBA is not the same as LBA, the controller is definitely important in IBM's transition to FBA. (Though it could perhaps better be characterized as a transition away from count-key-data, since they always had FBA devices on their smaller computers, though that preceeded their invention of the term “FBA”.) Controllers like the IBM 2841 could not handle FBA devices, with their different track structure.

The IBM 2311 model 11 did know about sectors: it provided a signal back to the controller whenever it passed the start point of a sector. While the cylinder and head part of the addresses were the same, the record number (on the 2311 model 1) was used by the controller to count records from the index point by interpreting the data stream from the device, whereas the sector number (on the 2311 model 11) was used by the controller to count sector pulses. John Sauter (talk) 13:57, 11 August 2014 (UTC)