[K12OSN] Just about ready to make a purchase

[Robert Arkiletian]

On 5/8/07, Jim Kronebusch <jim at winonacotter.org> wrote:
> I am finally ready to purchase the new server for our school and new thin clients.
> Below is what we'll be starting with, making the switch on labs and media centers first,
> next year adding teachers.
>
> -Dell PowerEdge 6800
> -Quad 3Ghz/800Mhz/4mb Cache Dual Core Intel Xeon 7130 Processors
> -16MB 400Mhz DDR2 RAM (8x2GB to start, will handle 64GB total)
> -Embedded PERC4e/Di RAID Controller
> -6 300GB 10K RPM Ultra 320 SCSI hard drives configured in RAID 10 (3 striped 300GB
> mirrors for a total of 900GB storage, machine will handle 10 SCSI drives and a 2 drive
> media bay for future expansion, from my research this will give me the fastest possible
> read/write speeds while maintaining full hot swap redundancy)

Your post stuck in my head this evening Jim. Few more points I wanted
to share. Those 8 cores are fed by (share) one 800Mhz FSB memory
controller using UMA (Uniform Memory Access). This is where Opteron
systems shine. Each Opteron has it's own memory controller. So you
would have 4 memory controllers controlling 4 memory banks. However, I
am not sure if NUMA (Non Uniform Memory Access) in the Linux kernel
works with dual core cpu's. Maybe someone can chime in if they know.
But I know 4 single core cpu's would enable NUMA support.

>From http://en.wikipedia.org/wiki/Non-Uniform_Memory_Access

"Now a system can starve several processors at the same time, notably
because only one processor can access memory at a time.

NUMA attempts to address this problem by providing separate memory for
each processor, avoiding the performance hit when several processors
attempt to address the same memory. For problems involving spread data
(common for servers and similar applications), NUMA can improve the
performance over a single shared memory by a factor of roughly the
number of processors (or separate memory banks)."

However, the shared memory portion of something like Firefox will
still probably use AMD's Hyper Transport bus.

BTW, 300GB scsi drives are awful expensive. If you are going to use
hardware RAID 10 have you considered enterprise SATA drives and a
dedicated PCI-X SATA controller with onboard write back cache (3ware
9550SX or LSI MegaRAID 300-8X). My guess is it would probably be
better than the controller that comes with the Dell. Maybe someone who
has used those controllers can comment.

Just checked the Dell PowerEdge 6800 specs and they have this under
the power supply section.

"Redundant power is available in all 200-240V configurations and most
100-119V / 120-127V configurations except those that contain or exceed
four processors, 32GB memory, five hard drives, and two PCI cards.
Redundant power availability will vary by configuration. NOTE: The
minimum configuration for most US based facilities is 120-127 Volts."

According to this, your setup probably requires too much power to
support redundancy.

-- 
Robert Arkiletian
Eric Hamber Secondary, Vancouver, Canada
Fl_TeacherTool http://www3.telus.net/public/robark/Fl_TeacherTool/
C++ GUI tutorial http://www3.telus.net/public/robark/