Re: Grind archive drive

[fmlin@netcom.com (Frank M. Lin)]

>>>Hi everyone,
>>>  I called a contact and got these prices for the drive.
>>>
>>>    Micropolis 1598  (5.25")	$720
>>>    Micropolis 2210  (3.5")	$950
>>
>>For Steve ( Chiang and Nelson )'s information, both are 1 gig drives.
>>( unformatted?! ), and a local HD vendor have the following data for them.
>>
>>1598 is the older series, which is 5.25" full-height.  Average access time
>>is 15ms.  2210 is a more more recent 3.5" half-height drive with average
>>access time of 10ms.  With the great price difference, I'd say go for
>>the 1598.  Since the access of the drive is over the network, I don't
>>think speed is that crucial.
>
>	Ahhh, ouch. Grind seems to becoming a rather popular spot, and
>the fact that the access is over the net makes a slower drive very bad.
>I'm not sure how Unix does it's disk buffering but I doubt if it will
>read whole files as soon as you ask for them.  The crucial point I'm
>trying to make is that if 10 people are d/l'ing 10 different files the
>disk is going to be getting rather thrashed.  Granted, these people's
>ftp will only be so fast, and I haven't done my math, but it seems that
>the 15 ms is going to be apparent a lot of the time, as the server takes
>care of each block to each person at a time.  Let's see, lets say the
>machine gets 10k at a time off the disk.  Let's say your ftp supports
>20k/sec.  This will all be thrown off if grind's net access is below a
>certain point. 20 people @ 20k/sec=400k/sec for the drive=

Secondary storages, like HDs are known to be slow.  Compare to the CPU and
memory.  I'm sure UNIX is smart to follow the basic rule, once we find the
data, copy everthing.  So I'm pretty sure it's moved to memory first.

These drive can do atleast 1mb/sec., and I'm sure 15ms will be sufficient.
I haven't heard Steve Nelson commenting on which drive he have in mind
though.  Anyone else have comments?

-- 
frank m. lin
fmlin@netcom.com