MQSeries: form an orderly queue within Advanced Peer-to-Peer Networking
CICS for OS/2 is not the only ‘middleware’ that IBM was touting at the end of March; the company continued the theme by unveiling the first implementations of its Message Queue Interface, dubbed the MQSeries, and said that it will try to promote them as a cross-system standard. Message Queue Interface first came to light in IBM’s Network Blueprint announcements last year (CI No 1,901) as one of a clutch of network interfaces supported within its Advanced Peer-to-Peer Networking architecture. It defines a way for applications on networked machines to pass information between themselves, and copes with one of the fundamental problems with distributed systems: namely that networks break and machines go down. The best way to explain Message Queue Interface is to compare it with a standard remote procedure call, where a client sends a request to another machine and waits for the response. With an RPC, if the remote machine is down then the request either has to be repeated or forgotten about, which makes writing predictable or robust applications difficult. By contrast, Message Queue Interface works in a time-independent fashion; requests are queued to be sent to the remote machine and if anything goes wrong they are automatically remembered and re-tried without the application having to do any work at all. The technique requires queue managers at either end and IBM says that it has these have been written for all of its operating systems, plus Digital Equipment Corp VAX/VMS, Stratus Computer Inc’s badged System/88 machines, and Tandem Computers Inc’s Guardian operating system.
MS-DOS and Windows
Even MS-DOS and Windows applications can use Message Queue Interface, but need a separate queue server running on an OS/2 machine. It is a remarkably simple idea, which IBM argues can act as a practical way of connecting existing islands of automation. The key point here is that it only attempts to link the islands into a loosely coupled way, and avoids the expensive development work needed to synchronise the two systems together exactly. Similarly, Message Queue Interface should improve the efficiency with which loosely coupled systems work: travel agents, for example should be able to request flight information from one system and get straight on filling in other details without having to wait for the remote system to respond. In the jargon, it means that applications can be ‘stateless’, that is, is the communication application having to keep track of what it state it was in when the failure occured. IBM is hoping that the popularity of workflow computing models will drive uptake, but it acknowledges that message queuing is not a new concept; what is new it says, is the attempt to implement it across multiple manufacturer’s machines. Unfortunately IBM has not submitted it yet to the likes of X/Open Co Ltd, neither has it signed any deals with the likes of DEC to push MQSeries on its own machines: instead the firm is relying on the system’s own merits to attract software developers.
RS/6000s pair off for fault-tolerance with High Availability Cluster
Clustering RS/6000 machines continues to hover near the top of IBM’s agenda – if you can really call two machines a cluster. The latest announcement in the field is for High Availability Cluster Multi-processing/6000 version 1.2: software that aims to glue a pair of the machines together in such a way as to combine the benefits of fault-tolerance and symmetric multiprocessing. Earlier releases of the software used two relatively simple ways to provide fault tolerance. Mode 1 or ‘standby mode’ left one machine idle, waiting for its partner – supporting all the users – to fall over and die, at which point it would cut in. Mode 2 (‘partitioned workload’) was slightly more sophisticated and enabled both processors to support their respective users and applications while providing mutual back-up. Now version 1.2 introduces Mode 3 or ‘loosely coupled multiprocessing’ where the two machines support the same users a
nd co-operatively run the same applications. On the fault-tolerance front IBM says that Mode 3 substantially cuts the delay between the one machine failing and the other one picking up where it left off; by the company’s own estimation the switch can take from anywhere between a long 30 seconds to a horribly long five minutes in modes 1 and 2. In mode 3, however any delay to end users should always be less than one minute says IBM; since both processors are already processing the same shared data. Using loosely coupled multiprocessing imposes a few limitations, both on programmers and hardware. In particular data being ussed by the dual processors must be stored on the the processors have to be attached to a IBM 9333 high performance disk system, shared data must be capable of storage on raw disk configurations rather than something like the the AIX Journaled File System and software must be specifically configured to work with the HACMP/6000 Distributed Lock Manager. IBM cites Oracle Version 7 Parallel Server as one example of a program that has been suitably tweaked – one copy is needed for each machine. For other developers there’s an application programming interface that gives access to the lock manager.
Feel like shouting at your micro? Voicetype will make it do something
Voicetype Control For Windows sounds like it may be a deliberate attempt to make Windows users unpopular with their colleagues it enables them to replace various keyboard and mouse commands with the spoken word. There is no indication of any OS/2 version. To run the program, users need a microphone an extra 2Mb of RAM, $130 to buy the software and a separate Sound Blaster audio board. The company is claiming voice-independent recognition of 64 commands at a go, and alternative vocabularies can be switched in as required. In addition the software can be trained to recognise its new master’s voice. IBM is usually pretty forthcoming about the markets that it forsees for its products, but in this case limits itself to saying that it will appeal to professionals who would prefer easy-to-remember voice commands to complicated keystrokes or mouse movements. It is much too politically correct to say that the one market to which it would be a real boon is the physically handicapped – in many degenerative diseases, the power of speech is one of the last to fade away.
