notes from last night

Steven Hanley sjh at
Fri Mar 1 10:26:17 EST 2002


I made some notes during the talks, I make no claims to their being useful or
accurate but hey here they are if people are intersted.

Burn Alting
intro to his more people do talks idea, then the why you would want to be at a 
dentist or what its like being a disk drive.
Works on a kernel for his company based on the plan9 system.
kept a dog theme by calling their kernel k9 (so as not to be sued), their
command interperator is called husky. Their stuff runs on mips, some of the
chipsets they have to work with suck a great deal.
They are a disk drive system, many disks in the back end one or so interfaces
on the front. Like one or two scsi interfaces as a front end for connecting to
the system.
Newer and bigger systems are now in use.
RAID because the market demands it, pointing out that there is no totally
redundant raid on this planet. More later.
Things that will fail in computers... just about everything.
because of the costs of getting data to allthe redundant points it is kind of
weird to have it everywhere, the firmware will die before other stuff?
adding more layers of parity etc is more expensive.
the operating system will issue the read or write request that actually hits
the disk, if an application issues a request it ddoesnt matter what size, the
OS will split it up into chunks it likes.
Can make disks work a lot better for you for a specific application if you cut
out the file system, do it yourself and you can get the utmost performance out
of it.
Any reset of the scsi bus tosses away any reservations.
The world is moving towards fibre channel for disk storage interfaces. (lower
level than scsi, well designed compared to scsi)
Much faster bus speeds, the bigger advantage is that it doesnt suffer from
some of the low level problems (burn described some of these, didnt note them
down) scsi does.
All the drive companies invested lots of money in fibre channel, due to this
serial ATA is not going to be out in big numbers for a while. Tooling up for
something new is expensive.
More accuratly, the amount of time invested in getting new technologies
running on all the server systems and the like, look how long fibre channel
has taken to hit main stream as an example of this "toolup" time.
Need to change your market, your engineers, etc is involved in this initial
To get as much performance out of an application find out exactly what an Os
is doing to the data as it goes to the disk.
No such thing as fault tolerant raid, the OS will crash before the board will
crash. A single DMA will be turned into three DMAs inside the fault tolerant
system, this is a big single point of failure.

Anti Roppola
FreeGIS talk
used to keep spatial location information for various objects. keep spatial
informatiuon about things in a database. Can then use this information for a
ehck of a lot of things. eg, tlstra keeps information of all their items
(telegraph poles, phone cabling, etc) in this sort of system. Then when thing
sfail they wont go out immediately to fix something, instead they wait until
there are a sufficient number of items needing work in one region at one time.
renders data to people, so it contains information aboout spatial relations
and people can ask to be given a map with various spatial data on the map, say
a map of .au with all the roads in the country on the map...
able to make queries to learn things abotu space. like wireless, query, if he
puts a wireless system in a given place what other nodes can it speak to
taking into account various information, such as hills, buildings, etc.
the interface and results for mapserver are all in vanilla html, there is no
excess javascript stuff as it is diffiicut to make it work across different
browser platforms
All of the stuff he is doing is server side, why they can get away with
vanilla html and such. They higher ups were surprised at the hardware they
were running on, using a pII 300, where some commercial soutions costing
insane amounts of money can not even be kept running for more than three days.
The map is stored as one huge tiff of .au to 250m scale, the system currently
just pulls that information out, combines it with the GIS data and displays it
in a simple png.
Each page on average is about 20KB (that his system sends out)

        See You

sjh at
You are subtle as a window pane standing in my view
but I will wait for it to rain so that I can see you
   Anticipate - Ani

More information about the linux mailing list