On Monday 15 April 2002 06:20 pm, Stephen Samuel wrote:
>
> It's stell cheaper to get 1~1.5G for 4 users than 512M for each of them,
> to avoid them spiking into swap space. Total RAM overhead is
> likely to be a bit less (shared memory), and memory or CPU
> usage spikes are easier to eat with a faster machine with a bit
> less ram than the total shared between 4 users.
>
> If a user 'spikes'' for a long period of time, than THAT ammount
> of ram should be considered the baseline for that user. In any case,
> It's still likely to be cheaper to buy the RAM needed to keep 4 users
> happy in one box than to keep them all happy in separate boxes.
It also makes sense to stick a cheap three or four disk IDE RAID in the box
and get some approximation of redundant data storage. (If you're sharing
everything but the user directories anyway, you can get away with devoting
1/4 the disk space to a parity disk without really losing out in bang for the
buck terms. AND you get more speed out of it (especially if you're
distributing swap space in paralell).)
You can even stick in a spare IDE controller in a PCI slot so each drive gets
to be a master on its own cable, to double the bandwidth again... (Sticking
a RAID in individual workstations, on the other hand, is probably expensive
overkill, and actually quadrupling the amount of maintenance since hard
drives are one of the main moving parts of the box. Distributing your swap
space will still crash the box when a drive dies, it just means your system
and data partitions should be easily recoverable when you reboot with a new
drive in there. We're not talking six nines of uptime for any box with only
one power supply anyway (ANOTHER moving part :). Although a UPS makes
economic sense for a shared box as well. (That that it could power four
monitors, but maybe four LCDs? Or at least save your data and shut down
cleanly. Swsup?)
A single 100baseT card for four users isn't going to be much of a bottleneck
(you can stream 70 simultaneous ~DVD quality mpeg4 video streams through ONE
of those cards), and you could save a lot of trouble on wiring too...
Sticking four users on a shared box at the intersection of four cubicles
seems quite doable to me. (Or one box per four four students in a university
computer lab environment.) If you're administering workstations for 100
people, it might not actually cut the workload by 1/4, but it still sounds
like a heck of an improvement.
Still needs rmap to enforce even remotely fair per-user swap behavior,
though... :)
Rob
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Mon Apr 15 2002 - 22:00:26 EST