Ultra 60 vs Ultra II

Ultra 60 vs Ultra II

Post by Charles Buec » Wed, 03 Jun 1998 04:00:00



Hi all,

I need to buy some new workstations and one server. Until now,
I was always going for Ultra 2 and Ultra 3000, but I'm giving a
look at Ultra 60 and 450 for the next buy, mostly for $$$ reason.

Some IS guy here told me the I/O performance of the new PCI
workstations (U30) is quite disapointing when compared to
a Ultra 1 / 170 he has, eg a tar extract took 3 minutes on U30
agains 2 on U1.

Can people around here comment on this issue, and maybee
give some good/bad feedbacks about their experience with
the new line (U30, U60 and 450) ?

What I will do is a new network management platform, with
one server used primary for data storage (NFS, maybee Oracle)
and user workstations running net mgmt tools (HP OpenView,
Cisco stuff for switching and so on). For now, we will have about
10 clients workstations, btu it may grew up to 30 in the next 3
years, and my buy should last this possible grow.

TIA, Charles
---------------------------------------------
Ch. Bueche / Unix, water, snow and Net surfer
---------------------------------------------

 
 
 

Ultra 60 vs Ultra II

Post by Francois Desjardin » Wed, 03 Jun 1998 04:00:00


Hi Charles,

We are using the 450 (server) and we are very please with it. You have
plenty of room for expansion: 4 CPU's, 20 slot for HD (now you can put
the new low profile 9.0 Gb), 3 Power Supplies. And for the price $, it
's great machine


> Hi all,

> I need to buy some new workstations and one server. Until now,
> I was always going for Ultra 2 and Ultra 3000, but I'm giving a
> look at Ultra 60 and 450 for the next buy, mostly for $$$ reason.

> Some IS guy here told me the I/O performance of the new PCI
> workstations (U30) is quite disapointing when compared to
> a Ultra 1 / 170 he has, eg a tar extract took 3 minutes on U30
> agains 2 on U1.

> Can people around here comment on this issue, and maybee
> give some good/bad feedbacks about their experience with
> the new line (U30, U60 and 450) ?

> What I will do is a new network management platform, with
> one server used primary for data storage (NFS, maybee Oracle)
> and user workstations running net mgmt tools (HP OpenView,
> Cisco stuff for switching and so on). For now, we will have about
> 10 clients workstations, btu it may grew up to 30 in the next 3
> years, and my buy should last this possible grow.

> TIA, Charles
> ---------------------------------------------
> Ch. Bueche / Unix, water, snow and Net surfer
> ---------------------------------------------


 
 
 

Ultra 60 vs Ultra II

Post by Jay Lesse » Wed, 03 Jun 1998 04:00:00




>Some IS guy here told me the I/O performance of the new PCI
>workstations (U30) is quite disapointing when compared to
>a Ultra 1 / 170 he has, eg a tar extract took 3 minutes on U30
>agains 2 on U1.

We just transitioned from U1/200 -> U30/300 for "standard desktop"
new purchases, and I had exactly this concern.

I set up a loaner demo U30 from the local Sun sales office; it had
exactly the same disk, same partitioning, and same memory as one
of my existing U1/170's.

I found that a "cptree" ("tar cf -" piped to "tar xf -") was faster on
the U30 by a factor of almost exactly 300/200, which mildly surprised
me.  I did several file I/O tests; in one case (single humongous file
copy) I found the U30 8% slower, in all other cases the U30 was the
same or faster.

CPU-bound tasks just scaled with clock speed more-or-less like you
would expect.

Go ahead with the U30's, they're good boxes.  We do find they generate a
godawful disk whine (same disk as our U1's, so this is just a packaging
difference).  One of my users who still has good high-frequency hearing
has improvised a packing-foam sound block, which works fine.  :-)

No experience with the 450.

--

Lattice Semiconductor Corp.                    (voice)1.503.681.0118
Hillsboro, OR, USA                               (fax)1.503.693.0540

 
 
 

Ultra 60 vs Ultra II

Post by Jon Reite » Wed, 03 Jun 1998 04:00:00


I've not seen that my Ultra 30 248 is slower than my Ultra 1 167.
Actually for every operation I've ever measured the 30 was at least as
fast.

The bandwidth to main memory is higher and the cache is larger so
memory-limited programs should be quite a bit faster.  This is partially
due to the higher clock rate but you can get 30s with higher clocks that
1s so this is _not_ an unfair comparison.

Most of my i/o operations are NFS not local disk so I can't really speak
to your friends claim per se, but I have would guess (s)he had a slower
disk in the 30 and it really isn't the _system's_ fault.

And since Sun is committed to PCI I can't imagine buying a SBus machine
unless it is really cheap or you need SBus because you already own 10000
SBus 622mb atm cards.

                                        -jon

 
 
 

Ultra 60 vs Ultra II

Post by William F. Wyat » Wed, 03 Jun 1998 04:00:00


[...]

Quote:> Some IS guy here told me the I/O performance of the new PCI
> workstations (U30) is quite disapointing when compared to
> a Ultra 1 / 170 he has, eg a tar extract took 3 minutes on U30
> agains 2 on U1.

[...]

What OS version was the IS guy running? On our U30, we found bad I/O
(specifically, disk I/O limited to 5 MB/sec) under the initial release
(2.5.1 3/97, I believe) that went away with the 8/97 release. I think
the PCI driver had some bottlenecks in the initial release.

--

   Smithsonian Astrophysical Observatory  (Cambridge, MA, USA)

 
 
 

Ultra 60 vs Ultra II

Post by Trevor Fiat » Thu, 04 Jun 1998 04:00:00




[chop]

Quote:>And since Sun is committed to PCI I can't imagine buying a SBus machine
>unless it is really cheap or you need SBus because you already own 10000
>SBus 622mb atm cards.

I'd say it's only a matter of time before you see an external SBus
card chassis with a PCI interface card.  I've heard that there are at
least two major Sun aftermarket parts makers working on them.

Once that's done, the schism will be complete and SBus will have both
feet in the grave...

-Trevor

--
/--------------------------------=------------------------------------\


\--------------------------------|------------------------------------/

 
 
 

Ultra 60 vs Ultra II

Post by Charles Buec » Thu, 18 Jun 1998 04:00:00





>>Some IS guy here told me the I/O performance of the new PCI
>>workstations (U30) is quite disapointing when compared to
>>a Ultra 1 / 170 he has, eg a tar extract took 3 minutes on U30
>>agains 2 on U1.

>What disks does he use?  I'm pretty happy with the
>I/O performance of my U-60.

Some new Seagate. I think he even mentionned he did use
the same disks on both machine when he saw the problem.

Quote:>(Or did he use NFS?)

No. Just some "tar x" on a local filesystem. I know dd is a bit
better.

Anyway, I will get 2 U60 next month so I will be able to
do my own homework. I wanted to go with the new (aka 30/60)
line anyway because IMHO the U2 won't last for long. Too bad.
It was a very nice design, albeit a bit expensive. I used to
manage a farm of about 40 boxes, 1/2 of them was U2 with
the storage expansion (wide scsi). The whole stuff was badly
fast when it went to i/o.

I'm now 5 years back, managing HP-UX 10.20. What a pain !
I should be back to modern UNIX soon :-)

Thanks Casper, Charles
---
"NT is enterprise ready, so Microsoft run its accounting on AS/400"
Charles Bueche, DM-PSO
Swisscom, Bern, Switzerland

 
 
 

Ultra 60 vs Ultra II

Post by Sam C. Nicholson ! » Thu, 18 Jun 1998 04:00:00



> Anyway, I will get 2 U60 next month so I will be able to
> do my own homework.

If you would perform the following test, I will add it to the
list I am maintaining.

Poke back inthe archives to find my earlier posts on this.

The test is to basically time a compile of GCC, with gcc as
the compiler and gnu make as the maker (to get serial and
parallel make times.)

You can grab my tarball at
        http://www.webrelay.net:/gcc-test.tar.{Z,gz}
and you can grab a precomiled gnu make at the same location
as /gmake

(I suppose that to be true to the GPL, I should tell you that
you can find the sources to these programs at:
        ftp:/prep.ai.mit.edu/pub/gnu )

Anyway, there is a script calles runtests in the top dir of
the result of the tarball that writes its outout in ../*log.

If you run the test, just email me the logfile, and tell me
the disk config on the test machine.

If you want to see the table so far:

        www.webrelay.net/gcc.compile.times

There is no legend in the file as yet.

regards,
-sam