[SunRescue] Speed of SS10 vs DEC 3000/300LX Alpha?

[User Bobkeys BSD Bob the old greybeard BSD freak]

> > It would be interesting to find out where the bottleneck is in the Alpha.
> > It looked like the whole machine was just slow, e.g., disk access was slow
> > apparent ethernet speed is slower, screen function is slower.  The little
> > SS10 really surprised me.
> > 
> > Any insights or thoughts on the differences are appreciated.
> 
> Got a few questions:
> 
> 1.  did you run X on the Alpha? obviously you didn't on the SS10

No.  Just a text based terminal, on the full screen console.  X is not
quite functional yet on the 3000 class machines.  Rumor has it that it
might be in the upcoming NetBSD release.

> 2.  did you compile a custom kernel for either machine?  dunno about NetBSD
> but OpenBSD's default filesystem cache = 5% of RAM. If you increase
> BUFCACHEPERCENT you get a filesystem boost.

Interestingly, the Alpha had a stripped custom config for just the 3000
class machines, stripped to the bare bonz.  The SS10 was running the
OpenBSD generic kernel.  No softupdates or anything exotic in either.

> 3.  speed of disks the same on each machine?

Both are 3600rpm things.  The alpha seemed slower in throughput to
the drives (RZ28's).  It seemed like throughput to disk took some
time longer on the alpha, possibly as if things like traversing
tree structure was slowing it down, which would be associated with
how the OS handled file system structure.

> 4.  were you on a switched ethernet connection or just a hub?

Private 10mb hub off a 10 foot fiber link to the main campus backbone
(a neat favor called in from the network boyz for installing their
main campus hubs in my closets).  I get 1mb through the hub between
machines and 600k out to the internet.  That link should be identical
identical to either machine.  They sit side by side on that hub.

I have noticed NetBSD taking a significant performance hit in the
1.5 release on things like my VAXentoyz.  I am beginning to wonder
if it might be due to sparc-specific tweaks generic in OpenBSD and
not in NetBSD.

I have a spare identical alpha at home.  I may load up the OpenBSD
port as opposed to the NetBSD port and compare that.  That would sort
out whether it was hardware specific or OS specific.

Are there any benchmarks that would test net throughput to disk
that I might try?

I just checked download speeds on remote files, and the sun had about
a 5% advantage, so I can't attribute the speed difference to ethernet
throughput.  It still feels like fs management overhead and disk i/o
speed diffs.

Move over Alpha... come on in SS10!   (:+}}...

Thanks

Bob