Cell Relay Archive

[crystal <crystal@postme.net>]

Impressive. 

I believe it's the 
	* 200Mhz P6 CPU; and 
	* 1GB 60 ns RAM
that allows the "buffer" between the 
	- FastEthernet controller and the 
	- dual-channel (33x2Mhz) SCSI cards.

Things can certainly be speeded up with 2 CPUs & increasing RAM (2 or
4GB) OR; using a mobo that has system bus speed beyond 33Mhz, and 64-bit
PCI.

Henry Yen <henryyen@netcom.com> wrote:
>  .1 Adaptec AHA-3940 PCI dual-channel narrow FastSCSI controller
>  .1 Adaptec AHA-3940UW PCI dual-channel wide UltraSCSI controller
>  .14 4.3GB fast-narrow SCSI drives (Quantum, Seagate)
>  .18 9.1GB ultra-wide SCSI drives (Micropolis, Quantum)
>  ...Each drive array is connected to its own Adaptec SCSI channel. 

With dual-channel, I guess its still 66Mbps max throughout.
How many bits are "narrow" & "wide" SCSI?
Just multiply that: (in bytes) X 
		    system bus speed (33Mhz) X 
		    2 (dual channel)
		    = max throughput.	
Theoretically this would be >100Mbps max throughput, but...

However harddisks are mechnical devices after all. The access speed (ms)
can't keep up wih the RAM access speed (ns). What is happening is the
CPU has to go through perhaps about 1000 interrupts-phases to match the
ms/ns difference in the read/write heads & the RAM electronics.

Even with DMA/33Mhz, where the CPU is freed from handling interrupts
from the SCSI controllers, it still has to "do something else" rather
than "checking interrupts" becuase the harddisk are still very much
mechanical devices.

That's why I think Solid State Disk (SSD) ICs are the next step for
secondary storage. This would have the potential of making secondary
storage almost like RAM.

Imagine such a machine having the true speed of perhaps brain neurons.
x00GB-x0TB is perhaps the size of a housefly's brain (or large beetle),
just enough for scouting robotic probes that have semi-intelligence.
Something like that seen in DSV SeaQuest "whiskers" semi-auto probes.    

I wonder what neuron-networks topology is best suited for these space
probes? Just like landing an army of insect-piloted probes on Mars. And
with MPP and LAN enclosed within a "whisker" probe chassis, a few
hundred of these aeronautical-or-land-or-underocean probes might do an
automatic survelliance of a region. 

Just like an intelligent child in a wilderness.

Best Regards,
Crystal.

Henry Yen wrote:
> the latest configuration blurb for ftp.cdrom.com:
> 
>  .wcarchive.cdrom.com is an Intel architecture PC machine 
>   running the FreeBSD
>  .operating system.
>  .
>  .Its configuration is as follows:
>  .
>  .Tyan 1662D multiprocessor motherboard
>  .      The motherboard has on it:
>  .              Natoma chipset
>  .              5 PCI slots
>  .              3 ISA slots
>  .              Two CPU sockets
>  .              8 SIMM slots
>  .              2 serial ports
>  .              1 parallel port
>  .              2 IDE interfaces
>  .One 200MHz P6 CPU ("Pentium Pro") w/512K L2 cache
>  .1GB of main memory (8 * 128MB 60ns EDO SIMMs)
>  .1 Adaptec AHA-3940 PCI dual-channel narrow FastSCSI controller
>  .1 Adaptec AHA-3940UW PCI dual-channel wide UltraSCSI controller
>  .1 Intel Pro/100B PCI 100Mbps Fast Ethernet controller
>  .1 StorQuest ISA SVGA adapter
>  .14 4.3GB fast-narrow SCSI drives (Quantum, Seagate)
>  .18 9.1GB ultra-wide SCSI drives (Micropolis, Quantum)
>  .1 Siliconrax SR-5910 rack mount computer chassis
>  .
>  .The disk drives are housed in 4 industrial rack mount 
>   chassis made by Kingston Technologies. Each drive array 
>   is connected to its own Adaptec SCSI channel. The IDE 
>   interfaces on the motherboard aren't currently being used.
>  .
>  .Our connection to the Internet is via 100Mbps Fast Ethernet 
>   that connects to core routers which in turn connect to 
>   most of the major interchange points on the Internet via 
>   multiple DS3 (45Mbps) and OC3 (155Mbps) circuits.
>  .Wcarchive is physically located at the CRL Network 
>   Operations Center in downtown San Francisco.