Advice: Upgrade from SS2 SS10

Don Gilbert gilbertd at chipmunk.bio.indiana.edu
Tue Mar 23 18:25:19 EST 1993


I replied to Joel privately a bit ago. Here I have a few additional 
comments for all.  The IU biology department just got a Sun Sparc 10/41
upgrade installed (this is the Supersparc that lots of people have
been griping at Sun about because they started selling it last May
but in many cases are just now delivering it).  The Sparc 10/41
is a fairly fast single RISC processor.  Benchmarks put it at about
3 x faster than the last generation, the Sparc 2.  My test with
fastDNAml bears that number out (fastDNAml is a sped up version of
Felsenstein's DNAml, produced by Gary Olsen & colleagues).

I would rate the upgrade from a Sparc 2 to a Sparc 10 (model 30 or 41)
as a good way to spend computing funds, if you do any thing on it
that requires speed (like phylogeny analyses, sequence searching &
matching, etc).  The model 10/30 is a bit slower than the model 10/41
(I haven't tested it), but still more than 2x faster than the Sparc 2.
I don't have any benchmarks or comments on the newer, low cost Sparcs,
the LX and Sparcstation Classic.  They are supposed to be about the
speed of the Sparc 2.

Please note that there can be software compatibility problems in moving
to Solaris 2 (for instance I use the public CAP software which doesn't
yet have a Solaris 2 version, and GCG software doesn't have a Solaris
2 version yet).  BUT, the Sparcstation models 10/20, 10/30 and 10/41
all run the older Sun OS 4.0.x (Solaris 1) fine.  That is what our
Sparc came with installed.

Some other reasons to upgrade:  The Sparc 10 supports multiple
CPUs fairly easily.  For another ~$6K I can approximately double
the computational capacity by dropping in another CPU (after I
upgrade to the current Solaris 2 os).   The Sparc 10 supports Fast SCSI
for about 2x faster disk access.  I can't speak for this
from experience, just from the literature.  This depends on using
the right disks and maybe a special card.

As Joel states, Sun is offering a Sparc 2 to Sparc 10 upgrade that
may expire sometime soon (like end of March?).

Here are some quick benchmarks w/ a biology application.  The "u"
time number is the important one.  At 10.3 secs, the sparc 10-41
is roughly 3x faster than the sparc 2.  The sparc 10-20 (a bit slower
than the sparc 10-30) is about 2x faster.   Subjectively, the
Sparc 10-20 console feels a good bit spritelier than the Sparc 2,
and the Sparc 10-41 is noticeably quicker than the Sparc 10-20.

fastDNAml timings.

time fastDNAml < testdata_6.phylip

 (sparc 2, cc compiler, 8 users,  load average: 0.70, 0.56, 0.22):
28.3u 0.7s 0:32 88% 0+572k 0+0io 0pf+0w

 (sgi iris 30, cc compiler, 2 users,  load average: 1.13, 1.12, 1.03)
18.6u 0.1s 0:30 61%

 (sparc 10-20, cc compiler, 1 user,  load average: 0.23, 0.19, 0.00):
14.7u 0.1s 0:17 84% 0+604k 0+4io 0pf+0w

 ( sparc 10-41, cc compiler, 1 user,  load average: 0.49, 0.38, 0.10):
10.3u 0.1s 0:12 84% 0+624k 5+5io 14pf+0w


In article <C45zE7.6Bu at ms.uky.edu> jmg at seqanal.mi.uky.edu (Joel Guthridge) writes:
>  I'd like some advice from the net about upgrading from SS2 running
>SunOS 4.1.1 to a SS10 (SparcStation 10, model 30) running Solaris 2.1.
>The SS2 mainly runs a Molecular Biology Server, with the IG Suite,
>some public domain programs, a gopher server etc.  What are the advantages/
>disadvantages to upgrading?  For the molecular biol. programs we run
>will the SS10 cause a dramatic increase in speed.  My impression is that
>the main applications used (fasta etc.) are mainly limited by disk I/O
>speed, so the SS10 wouldn't increase speed of those applications much.
>What problems can you forsee in carrying out such an upgrade, ie. getting
>things going like they are now on the SS2?
>
>  The reason for asking is that Sun is offering an upgrade, for what
>appears to be a reasonable price, and we actually have the money to do it.
>I'm just not convienced that this upgrade is necessary at this time.  Maybe
>I should spend the money on other periphs., like diskspace, tape backup etc.
>
>
>   Thanks for the input!
>
>   Joel Guthridge
>   email:  jmg at seqanal.mi.uky.edu


-- 
Don Gilbert                                     gilbert at bio.indiana.edu
biocomputing office, biology dept., indiana univ., bloomington, in 47405




More information about the Bio-soft mailing list