Re: LX164 -IDE problems - thanks

["T. Weyergraf" <kirk@colinet.de>]

Hi,


> Just curious, why the extra controller?  My experience is that even an 
> Abit BP6 with two 466 MHz Celerons and software RAID0 and 5 gives great 
> NFS performance, even with public Debian and kernel mirrors.  (I'm 
> pretty sure it saturates 100 mb ethernet with NFS, because Debian 
> upgrades via NFS on machines across the same switch are very much faster 
> than on machines in another room across MITnet with a 10 mb drop.)  I've 
> never seen any reason to put in an extra controller -- unless you plan 
> to put in more than 4 IDE devices.

You are absolutely right, normally. However, there are some pretty good
reasons, why I plan on using a seperate controller:

1. the NFS-network is planned to receive a 1Gbit-ethernet fairly soon.
2. I cannot rule out, that therer will be 4 NFS-drives soon, plus a seperate
    of which the NFS-server itself boots
3. The HDD's i will most likely use, have rather large read-caches, since
    it can be expected, that there will be "burst-type" traffic from multiple
    machines on the same - quite large - files. The buffercache will take care
    of most of the impact, but OTOH, the premium for large-cache drives is
    small, as well as the second controller. Thus, my preference goes for
    UDMA-100 interfaces.

I *will* do some performance testing prior to buying the stuff - as far as test-drive
availability goes ( the local dealer, where i purchase my stuff, allows me to use
some of the material he uses for testing himself - quit nice, IMHO ).

I would prefer a nice 0+1 scsi array with noatime mounts and reads to the
array serviced by the mirror-set in parallel, but that is *way* out of my budged.

Anyway, thanks for your comments

Regards,
T. Weyergraf

-- 
Thomas Weyergraf                                                kirk@colinet.de
My Favorite IA64 Opcode-guess ( see arch/ia64/lib/memset.S )
"br.ret.spnt.few" - got back from getting beer, did not spend a lot.