The following message is a courtesy copy of an article that has been posted to bit.listserv.ibm-main,alt.folklore.computers as well.
Morten Reistad <[EMAIL PROTECTED]> writes: > Also, log structured file systems, the jfs and contributions to efs3, > and huge improvements to the irq and dma routing; including some work > in processor affinities. metadata logging is slightly different from log structured file systems. one of the problems with log structured file systems is the periodic "garbage collection" done to consolidate files, making their records sequential and contiguous. for other drift ... during work on HA/CMP http://www.garlic.com/~lynn/subtopic.html#hacmp we hired one of the people responsible for doing the BSD log structured filesystem implement to consult on doing a "geographically distributed filesystem". JFS was originally done by people working on 801/AIXV3. 801 early on had definition/implementation for "database memory" ... i.e. hardware could keep track of fine-grain changes (size on the order of cache-lines). Just load up data into memory mapped infrastructure ... provide the COMMIT boundaries ... and eliminate needing to sprinkle "log" calls thruout the code. At commit, just run thru the changed memory indications ... collecting data-lines needing logging. There had been various kinds of conflict between the unix development group in palo alto and the group in austin. The palo alto group took JFS and ported it to non-801 platforms ... having to retrofit the logging calls to the software (since they lacked database memory hardware). It turns out that the version with explicit logging calls ran faster than the original implementation (even on the same 801 hardware platform) ... the commit time scanning of memory for changes tended to be higher overhead than the explicit log calls. Then the remaining justification for database memory is the implementation simplification ... somewhat akin to some of the pushes for parallel programming (except parallel programming is frequently explicitly about performance; not trying to trade-off performance against simplicity). some of the database memory stuff can be found under the heading of transactional memory ... some posts mentioning transactional memory: http://www.garlic.com/~lynn/2005r.html#27 transactional memory question http://www.garlic.com/~lynn/2005s.html#33 Power5 and Cell, new issue of IBM Journal of R&D http://www.garlic.com/~lynn/2007b.html#44 Why so little parallelism? misc. past posts mentioning log structured filesystems http://www.garlic.com/~lynn/93.html#28 Log Structured filesystems -- think twice http://www.garlic.com/~lynn/93.html#29 Log Structured filesystems -- think twice http://www.garlic.com/~lynn/2000c.html#24 Hard disks, one year ago today http://www.garlic.com/~lynn/2001f.html#59 JFSes: are they really needed? http://www.garlic.com/~lynn/2002b.html#20 index searching http://www.garlic.com/~lynn/2002l.html#36 Do any architectures use instruction count instead of timer http://www.garlic.com/~lynn/2003b.html#69 Disk drives as commodities. Was Re: Yamhill http://www.garlic.com/~lynn/2004g.html#22 Infiniband - practicalities for small clusters http://www.garlic.com/~lynn/2005l.html#41 25% Pageds utilization on 3390-09? http://www.garlic.com/~lynn/2005n.html#36 Code density and performance? http://www.garlic.com/~lynn/2006j.html#3 virtual memory http://www.garlic.com/~lynn/2006j.html#10 The Chant of the Trolloc Hordes http://www.garlic.com/~lynn/2007.html#30 V2X2 vs. Shark (SnapShot v. FlashCopy) http://www.garlic.com/~lynn/2007i.html#27 John W. Backus, 82, Fortran developer, dies some past posts mentioning database memory http://www.garlic.com/~lynn/2002b.html#33 Does it support "Journaling"? http://www.garlic.com/~lynn/2002b.html#34 Does it support "Journaling"? http://www.garlic.com/~lynn/2003c.html#49 Filesystems http://www.garlic.com/~lynn/2003d.html#54 Filesystems http://www.garlic.com/~lynn/2005n.html#20 Why? (Was: US Military Dead during Iraq War http://www.garlic.com/~lynn/2005n.html#32 Why? (Was: US Military Dead during Iraq War http://www.garlic.com/~lynn/2006o.html#26 Cache-Size vs Performance http://www.garlic.com/~lynn/2006y.html#36 Multiple mappings http://www.garlic.com/~lynn/2007i.html#27 John W. Backus, 82, Fortran developer, dies ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html