[Info-vax] The Future of Server Hardware?

[Paul Sture]

In article <5069f654$0$1247$c3e8da3$b1356c67 at news.astraweb.com>,
 JF Mezei <jfmezei.spamnot at vaxination.ca> wrote:

> On 12-10-01 15:21, David Froble wrote:
> 
> > Weren't racks of a "gazillion" DS10Ls a good solution for some problems? 
> >   It seems that plenty of them were made and sold.
> 
> 
> If a rackful of DS10Ls was the best solution, why then did people buy
> Wildfire/Galaxy class machines ? 

In our case it was a relatively simple upgrade path from existing 8400 
class systems using existing apps on VMS.  Risk was a big factor here 
because a lot of money could have been lost with production downtime.

> In the case of Google or facebook, do they really need that much CPU
> power, or is it IO bandwidth that they need ?
> 
> And if Google stopped using gratuitous HTTPS for public data such as
> youtube etc, how much spare CPU would they suddently have in their data
> centres ?

One data point here.  Yesterday I used SCP to transfer a 250GB file from 
one system to another.  On the receiving end (a 1.4 GHz dual core AMD) I 
saw both processors at nearly 100% utilisation, and on the sending end 
(a 2.9 GHz dual core AMD) one processor was at 100%, the other at about 
80%.  The operating systems were CentOS and openSUSE respectively. I got 
a transfer speed of somewhere between 30 and 40 MB/s.

Both systems gave adequate response times for other stuff I was doing at 
the time.

-- 
Paul Sture