[rescue] New Niagara Sun boxes... useful link for quick specs

velociraptor velociraptor at gmail.com
Thu Dec 8 08:36:55 CST 2005


On 12/7/05, Patrick Giagnocavo 717-201-3366 <patrick at zill.net> wrote:
> I came across this:
>
> http://blogs.sun.com/roller/page/paulj#niagara
>
> which gives a basic rundown of the new Sun CPU's capabilities.
>
> It looks good, especially since they implemented common crypto operations in
hardware, so supposedly you could for instance, encrypt everything going
into/out of your system with minimal hardware slowdown.
>
> On my moderately-loaded v20z, I see:
>
> Total: 498 processes, 1563 lwps, load averages: 0.07, 0.06, 0.05
>
> so if there was a lot of activity across a number of apps, I could see that
performance on a "threaded" CPU like Niagara might be pretty good.

For our Solaris-based firewalls, this would be a huge improvement.

Observing their performance, I noted that they were doing 3K+ context
switches a second.  2-way 900Mhz  is not cutting it.

The sad thing is the person in charge of them (sr. network engineer)
was looking at iostat and b/c the three cpu columns added up to 100%,
assumed the were "doing ok".  He didn't even use "uptime" to note that
the load was 15+.  They are being replaced with 4-way boxes, and b
umped to Solaris 9.  But I fear that will not be enough as we are getting
a newer version of the firewall software as well, and we all know how
COTS generally blows^H^Hats.

I don't think a lot of people actually *look* at their boxes and figure out
why they are hitting performance issues.  They just tell management,
"we need a bigger box", because that's easier to explain to wintel/mhz
mentalities than, "we are CPU bound" we need something 4- or 8-way,
etc.

I'm just glad to see that Sun has figured out that price is an issue and
are doing something about it.  I'm still worried about the life of the
sparc line, but this should resuccitate it some.

=Nadine=



More information about the rescue mailing list