"PALO ALTO, Calif. (AP) -- Sun Microsystems Inc. is requiring its 38,000
U.S. workers to take the first week in July off, part of efforts to trim
costs amid a slump in demand for its high-end workstations and servers."
- from the Associated Press news service (April 25, 2001)
"We put the dot in dot com."
- from a Sun Microsystems ad campaign two years ago
( ESNUG 369 Subjects ) ------------------------------------------- [04/26/01]
Item 1 : Sun PR Troubles, The Official Sun Reply & "Easy Setup HP-UX 11.0"
Item 2 : Cooley's An Idiot -- We've Seen No Such Troubles With *Our* Suns
Item 3 : Hey, John, We Saw Similar Problems With Our Sun Workstations, Too!
Item 4 : Linux PC's Are Cheaper! Why Not Use Them Instead Of Pricey Suns??
Item 5 : Customers Benchmark EDA Tools On Linux PCs vs. Sun Workstations
Item 6 : Sun's Supposedly "Open" Culture & Cooley's Old/New Coke Remark
Item 7 : But Needing 64-bits Kills Any Serious EDA Migration To Linux PCs
Item 8 : Yes! We Also Got Mentor To Port Its EDA Tools To Linux, Too!
Item 9 : O'Reilly's Linux Clustering Book & Google Uses 6,000 Linux Boxes
Item 10 : Hmmm... Synopsys Ported Everything *But* EPIC To Linux... Hmmm...
Item 11 : Intel?, Athlon?, Abit?, Asus? -- What Is In NVidia's Linux Farm?
Item 12 : Simon Read -- Linux On SPARCs & Beowulf Linux Network Clusters
The complete, searchable ESNUG Archive Site is at http://www.DeepChip.com
( ESNUG 369 Item 1 ) --------------------------------------------- [04/26/01]
Subject: Sun PR Troubles, The Official Sun Reply & "Easy Setup HP-UX 11.0"
> The way I found out about it was in an e-mail from Jon Stahl of Avici. "I
> was wondering if I could poll the ESNUG community for recent experiences
> regarding the use of HP workstations instead of Sun as a platform for
> EDA," asked Jon. "The reason I bring this up is that we have had a
> ridiculous amount of trouble with our Sun hardware and software over the
> last few years, and we are at the point of researching alternatives."
>
> - from "Sun Reveals A New Dark Side"
From: "Pat Duba" <pat_duba@hp.com>
Hello John,
I was passed your posting (as follows). Just wanted to let you know that HP
realizes the opportunity and the challenge that we have here. We're working
to make HP-UX easier to setup in a Solaris environment. We just started
shipping a new configuration "Easy Setup HP-UX 11.0" (HP p/n B5532A)
and working on other things, such as field resources, to make the setup of
HP-UX systems in Solaris environments go smoothly. We are shipping you an
easy setup media kit and User Guides. We have also "started" a web site
resource www.hp.com/go/easy-setup to help our users.
- Pat Duba
Hewlett-Packard Ft. Collins, CO
---- ---- ---- ---- ---- ---- ----
From: Abbe Patterson <abbe@purpleyogi.com>
John,
Great comments on Sun and I agree whole heartedly! There will come a day
when companies view strategic communications as more than hiring PR flacks
to work for marketing. Microsoft made such an effort in creating a position
for Linda Stone -- but Microsoft has a LOT of work yet to do on the
communications front. Sun lost some very senior, very good communications
folks over the last 18 months or so. It shows in recent "damage control"
efforts...
- Abbe Patterson
Skyblaze Ventures, LLC Santa Clara, CA
---- ---- ---- ---- ---- ---- ----
From: Dennis Kelly <Dennis.Kelly@Sun.COM>
John,
Here's Sun official statement on the "E-cache" issue:
Our messages on this issue remain the same:
-- This problem affects a very small number of customers, so small that
we are able to work with them on a one-to-one basis.
-- We have taken an aggressive approach to these problems and have a
number of solutions available for our customers.
-- We have, and will continue to, communicate forthrightly with our
customers.
Sun Microsystems has been working for almost a year with a small number
of customers who have experienced component quality problems. These
components issues mean that under, certain conditions, certain systems
reboot. There is no data corruption involved with these intermittent
problems. This is not new news to our customers.
Sun has taken an aggressive approach to addressing these issues with our
customers. We have identified the source of the problem and taken steps
to ensure that it does not happen again. We have communicated
forthrightly with our customers, and will continue to do so.
End of official statement.
Thanks,
- Dennis Kelly
Sun Microsystems
---- ---- ---- ---- ---- ---- ----
From: "Pat Duba" <pat_duba@hp.com>
Hello John,
I had sent you an email earlier today mentioning that we would send you
an Easy Setup HP-UX 11.0 kit. If you want one could you let me know
your shipping address?
- Pat Duba
Hewlett-Packard Ft. Collins, CO
---- ---- ---- ---- ---- ---- ----
From: Dennis Kelly <dennis.kelly@sun.com>
John,
Your bashing of Sun ("Sun Reveals A New Dark Side") made me think about
how best to respond, other than the Sun party line which I did twice. One
of my colleges came up with this proposal, which I'll support. John, I'll
loan you a Sun workstation, you try to break it (with EDA SW exercises,
not a ball peen hammer.) Put it side by side to an Intel box, or your
favorite HP-UX box, and report the results back in 4 months.
What do you say Gadfly?
- Dennis Kelly
Sun Microsystems
( ESNUG 369 Item 2 ) --------------------------------------------- [04/26/01]
Subject: Cooley's An Idiot -- We've Seen No Such Troubles With *Our* Suns
> Apparently, the external memory caches on Sun's Ultra servers were running
> into parity problems, causing spontaneous crashes. This was affecting a
> lot of Sun customers. "Paul McGuckin, an analyst with Gartner Group who
> deals regularly with major corporate customers, said that roughly 60 major
> Gartner clients have reported problems with as many as several hundred Sun
> servers," reported the Journal article.
>
> - from "Sun Reveals A New Dark Side"
From: Bill Rodewald <bill@eda.mke.ab.com>
Hi John,
I've been a Sun loyalist for quite some time and I'm sorry they didn't come
clean on their cache problems. I haven't seen any of the problems myself
(probably because we are using inexpensive Ultra10s and older Ultra1s and 2s),
but the problem seems to be documented at:
http://www.theregister.co.uk/content/2/17417.html
I've worked with Suns and older HPs, and a) Suns have been very reliable,
and b) the Sun bang-for-the-buck has been very good. I'd be interested in
any stories you come across on Sun's low end Sunblade 100. List starts at
$995, its only 500Mhz but it uses cheap SDRAM (up to 2Gb) & IDE hard disks.
- Bill Rodewald
Allen-Bradley Milwaukee, WI
---- ---- ---- ---- ---- ---- ----
From: Thomas Loftus <tloftus@intrinsix.com>
John,
At my previous employer we had many Sun Ultra-30, Ultra-60, and E-450
machines for doing ASIC design. We never had any trouble like is being
described. They would stay up for hundreds of days doing ASIC simulation
and synthesis. By the way, we generally used third party memory as well
(like Dataram or Kingston).
At my current job, we continue to use dual-CPU Ultra-60's and Ultra-80's
with 2-4 GB memory without any trouble.
Which models of Ultra Servers are you referring to? Is it operating
system/patch dependent? Does anyone know what the issue really is and on
which platforms it is occurring?
For future upgrades, I guess I would like to know more of the details on
this issue rather than a general "Sun hardware stinks."
I would also point out that at my previous employer we used a large number
of HP machines as well without much trouble. Having both was occasionally
an administrative burden, but our management felt it kept the workstation
sales people on their toes.
- Tom Loftus
Intrinsix
---- ---- ---- ---- ---- ---- ----
From: Steven Gelsie <steven.gelsie@jhuapl.edu>
John,
We have been using Sun's for years ( mainly workstations and smaller
servers) and have had very little problems (no more then HP's). The
memory problem you mentioned has been know for several months. Sun
seems somewhat negligent in making users aware of this memory problem
in their larger servers. The argument for Sun's is that they have a
better kernel, more software, and are more progressive.
The only place I thing HP's are better then Sun is Admin tools and
printing.
- Steve Gelsie
Johns Hopkins Univ. Laurel, MD
---- ---- ---- ---- ---- ---- ----
From: Romas Rudis <rudis@intrinsix.com>
John,
If what you say is really true, I agree this definately is the wrong
approach for Sun.
At Intrinsix we have all sorts of Ultra machines from Sun without seeing
any of these problems: Ultra 1, Ultra 2, Ultra 5, Ultra 10, Ultra 60,
Ultra 80 (including Tatung versions). The new Sun Blade 1000
(UltraSparc III) seems to work fine, too.
I guess we must just be very lucky...
- Romas Rudis
Intrinsix
---- ---- ---- ---- ---- ---- ----
From: Steven Gelsie <steven.gelsie@jhuapl.edu>
John,
Let's face it: Solaris is a much more robust and easier to maintain system.
Have you every tried to tuning the kernel on a Linux box? What's with this
Sun Solaris bashing? Go bash Microsoft if you want some one to bash.
You're losing your credibility. Capability is more important then speed
at doing one thing.
- Steven Gelsie
Johns Hopkins Univ. Laurel, MD
( ESNUG 369 Item 3 ) --------------------------------------------- [04/26/01]
Subject: Hey, John, We Saw Similar Problems With Our Sun Workstations, Too!
> Apparently, the external memory caches on Sun's Ultra servers were running
> into parity problems, causing spontaneous crashes. This was affecting a
> lot of Sun customers. "Paul McGuckin, an analyst with Gartner Group who
> deals regularly with major corporate customers, said that roughly 60 major
> Gartner clients have reported problems with as many as several hundred Sun
> servers," reported the Journal article.
>
> - from "Sun Reveals A New Dark Side"
From: "Ray Livingston" <rayl@atc.creative.com>
John,
For whatever it's worth, at my last company we bought an E4500 with 8 CPUs.
During the first 3 months, we had 7 CPU replacements (involving 3 slots on
2 trays) due to cache parity errors. The fix was to replace the trays, NOT
the CPUs. Sun was, however, always responsive, and never asked us to keep
it quiet.
- Ray Livingston
Creative Labs, Inc.
---- ---- ---- ---- ---- ---- ----
From: Denise Johnson <denise@berkeley.atmel.com>
Hi John
I read your article "Sun Reveals A New Dark Side" and would like to report
that last year (Jan 2000) we upgraded our network and purchased a Sun A1000
managed by an Ultra5. We are a small R&D team in Atmel of about 15
engineers, 30 to 40 workstations, PCs and Suns. Since the beginning of
this year we have had periodic soft hangs for no apparent reason -- usually
over the weekends when no one is running anything, the server will just
freeze up.
Sun Tech Support first recommended upgrading the memory, which we did, max
at 512 MB. And of course installing all the latest patches. But it keeps
happening. Each time they find another patch that is not current. Here is
the most recent suggestions from Sun:
Denise,
I think there is a memory leak on your system which could explain why
it stayed up&running longer after adding physical memory.
The kernel size was 198Mb when the system hung that corresponds
at 40.09% of the maximum size (~40% is too much). Most of this
memory is used by 2 kernel caches:
streams_dblk_1920 (63Mb) and ufs_inode_cache (20.9Mb)
The size of ufs_inode_cache looks correct for your system but the size
of streams_dblk_1920 is big which could cause your system hang. Most of
the time this buffer is used by drivers like hme, ip, tcp, ...
Your hme driver is not patched correctly (108263-01), the last revision
of this patch is 108263-06. It seems that you are using SNMP on this
system. SNMP driver needs to be up to date (last revision 107709-07),
it is source of memory leak.
Be sure to install these patches in single-user mode and reboot
the system after installation done.
After installing these patches, it would be necessary to monitor the size
of streams_dblk_1920. To do it, we need to set the kmem flags enabled in
the kernel. On Solaris 7 (64-bit), it can be enable by placing the
following entry in the /etc/system file, and reboot
set kmem_flags=0x3f
Then monitor the streams_dblk_1920 cache:
# echo kmastat | crash
Wait for the "memory in use" column of streams_dblk_1920 to grow to over
40 Megabyte, and then run:
# echo "kmausers streams_dblk_1920" | crash
and send us the output. We might need them to force a kernel dump
depending on whether we can determine the cause of the leak with only the
kmausers output.
If the size of streams_dblk_1920 never goes over this value, it means
that the patches fixed the problem and the kmem flags can be disabled
(removing the line in /etc/system and reboot).
- [ Sun Support Engineer Name Deleted ]
Now my questions to you, could we be having the memory cache problem you
are talking about in your article? And if so who should be talk to?
We greatly appreciate any information.
- Denise Johnson
Atmel Corporation Berkeley, CA
---- ---- ---- ---- ---- ---- ----
From: Jens-Uwe Walther <Jens-Uwe.Walther@force.de>
John,
A collegue of mine just sent me your article "Sun Reveals A New Dark Side".
We have this hardware problem since October last year. We bought about 40
Ultra 5 workstations. And until now I guess about 25 of the 40 had
hardware errors like "Red State Exception" and "Ecache Data Parity Error"
even for machines coming fresh out of the box.
The Ecache problem should be fixed now with the newest kerlnel patches
which address this problem. But I'm not sure if this finally solves the
problem. I patched all the machines due to this and now the floppy doesn't
work correctly anymore (you don't see files you've written on a PC to a
floppy and vice versa).
This is very annoying because it takes much time making calls to the Sun
service, prepare spare workstations for the Engineers and so on. But until
now I haven't have contacted Sun. Maybe I should. But it is good to know
that we are not alone with these problems.
- Jens-Uwe Walther
Force Computers Germany
---- ---- ---- ---- ---- ---- ----
From: [ Chicken Little ]
I saw your writeup about Sun's dark side. Their memory problems were
supposedly responsible for the EBay's outages last year.
Sun is one of the very few electronic design companies that don't use
Denali's memory models in their simulations. Ask them yourself!
Please keep me anonymous.
- [ Chicken Little ]
---- ---- ---- ---- ---- ---- ----
From: Richard_J_McWaters@raytheon.com
I got a new Sun Ultra 3500 server this past week. Brand new, out of the box
it had a bad SCSI controller and a bad disk controller. It took a week to
get a brand new machine fixed. The only reason we use Sun over HP is
"sticker price". My management feels that cheaper is better. In the case
of Sun, you get what you pay for.
My old 3000 server ran into some problems last year with memory. When they
went to change the memory chips on the CPU card, the heat sinks on most of
the hot components fell off. Good quality. I have had this system for
about 3 years, and about a year into my leasing contract I had a problem
with the RAID cabinet. The service guy who came out told me this big
expensive RAID cabinet was obsolete!
Sun has a history of "doing it's own thing". While their "open" policy on
Java sounded like a good thing, in practice, they have always had a it was
Not Invented Here attitude about all their hardware and software. I expect
that you will continue to be disappointed in their stuff.
Having had experience with IBM, HP, Apollo (remember them), DEC and Sun,
HP is the only one that consistantly produces quality hardware and software.
They are also the most expensive of the group, and therefore loose out to
the "buck cheaper" crowd.
- Richard McWaters
Raytheon
( ESNUG 369 Item 4 ) --------------------------------------------- [04/26/01]
From: Ross Smith <ross.smith@theseus.com>
Subject: Linux PC's Are Cheaper! Why Not Use Them Instead Of Pricey Suns??
John,
Chris Malachowsky's presentation at SNUG this week indicates that what Sun
does may be irrelevant. The number of Suns in Nvidia's compute farm has
remained constant over the past few years, while the number of Linux-based
PCs has grown exponentially.
- Ross Smith
Theseus Logic, Inc. Sunnyvale, CA
---- ---- ---- ---- ---- ---- ----
From: [ The Knight Who Says "Ni" ]
Hi John,
Keep me anonymous on this one please.
Linux is moving up as an alternative. I've been using VCS, Finsim and
Signalscan under Linux for a few years. Lately I've also been using
Cadence Verilog-XL under Linux. I have to say that I'm very happy with
the performance of these tools under Linux. I can also buy ten 1 GHz
Athlon systems for the price of one SunBlade.
I just wish more tools were available under Linux, e.g. Design Compiler,
FPGA Compiler II, PrimeTime etc. Some will be available soon according
to Synopsys (anybody beta testing out there?).
I would like to hear other peoples experience with EDA software under
Linux. According to an earlier servey posted here 9% of all designers
used Linux.
- [ The Knight Who Says "Ni" ]
---- ---- ---- ---- ---- ---- ----
From: "Erik Jessen" <erik.jessen@home.com>
John,
I get get the distinct impression I should be investing in RedHat...
I look forward to seeing how Sun responds to your ESNUG.
- Erik Jessen
( ESNUG 369 Item 5 ) --------------------------------------------- [04/26/01]
From: Alain Raynaud <alain@tensilica.com>
Subject: Customers Benchmark EDA Tools On Linux PCs vs. Sun Workstations
John,
This topic should interest your readers. It's information I had been
wanting to hear for quite a while, and now I have it!
Should we switch our simulation jobs to a farm of fast, cheap PCs?
I benchmarked a 1 GHz PC, fully loaded, running Linux, against one of
our Sun servers (400 MHz UltraSparc). The results? Not what I expected.
For VHDL simulation (using a leading VHDL simulator), the PC was 25%
faster than the Sparc.
For Verilog simulation (again, with a leading Verilog simulator), the PC
and the Sparc had exactly the same performance.
So what is going on here?
I have done other measurements with 600 MHz PCs, and the results scale as
you would expect. Keep in mind that I'm testing our RTL design only, so it
may not be your typical design. But that's what we care about. Cache size
and memory subsystem performance may matter more than pure MHz numbers.
In the interest of full disclosure, there is one job where the PC was 40%
faster than the Sparc. That confused me even more.
Also, this benchmark also gave me the opportunity to compare simulation
speed for the exact same design written in VHDL and Verilog. While I can't
publish the detailed results, I can confirm that VHDL simulation is slower
than Verilog.
- Alain Raynaud
Tensilica, Inc. Santa Clara, CA
---- ---- ---- ---- ---- ---- ----
From: [ Elbow Grease ]
Hi John,
Please do not use my or my company's name.
We did preliminary experiments comparing the performance of EDA SW; notably
simulation, STA, and DC / PhysOpt synthesis (beta code). We compared Sun
Solaris servers (~$40-60K) to 800MHz PC's running Linux.
Surpisingly, the Linux PCs outperformed the Sun boxes by anywhere from
50%-200%. It is too early to draw solid conclusions, however it raises an
interesting question of whether one might be able to get 2x the throughput
for a design task for less money on the HW and O.S. These were real
designs, by the way, not tiny university benchmarks.
- [ Elbow Grease ]
( ESNUG 369 Item 6 ) --------------------------------------------- [04/26/01]
Subject: Sun's Supposedly "Open" Culture & Cooley's Old/New Coke Remark
> Where the norms set by Microsoft, IBM and Intel are of paranoia, control
> and secrecy -- Sun has historically been a breath of fresh air with its
> openness and honesty. You can use Java because Sun open-sourced it. Sun
> will even give you the source code for StarOffice and Solaris. Openness
> is one of the keys to Sun's corporate culture.
>
> - from "Sun Reveals A New Dark Side"
From: Matt Staben <matt@monaco.com>
John,
You mention that Sun was perceived a "breath of fresh air of openness and
honesty." I find this a bit naive because period of time when the Open
Software Foundation was establishing its roots, the "other" meaning for
OSF was often said to be "Oppose Sun Forever." So, when using the Alpha
OSF/1 workstation, for example, remember that the OSF was working towards
what you believe Sun to have been.
- Matt Staben
Monaco.com
---- ---- ---- ---- ---- ---- ----
From: Tom Sharpless <tomsharp@netaxs.com>
Come on, John;
1) Sun built its business by selling hotrodded hardware of sometimes
doubtful reliability. This is not the first or 2nd or 3rd time
they've had to deal with shipped systems that flake out. What may
be new is that they are worried about publicity. That's part of
the price for selling "corporate iron", and it has nothing to do
with technical openness.
2) However IMHO, Solaris is as proprietary a Unix as you'll find.
Sun successfully sells stuff that's free on every other Unix
(e.g: you have to buy a compiler license from Sun to get grep
and other basic utilities). And they are surely not above tweeking
the system to lock in lucrative application markets. With Sun,
open source is pretty much a public relations stance -- just try
building you own Solaris (or Java VM, or whatever) from Sun source!
3) So let's not bewail a Sun that never was. They have always
been greedy hustlers, its just that once they had something to sell
that you couldn't get anywhere else.
Best regards,
- Tom Sharpless
Netaxs.com
---- ---- ---- ---- ---- ---- ----
> Most MBA programs teach the "New Coke" story as a cautionary tale about
> not fixing what ain't broke. I'm hoping that the management at Sun has
> learned that suddenly adopting a new closed-door, adversarial corporate
> culture during a crisis isn't good for business. And they should
> occasionally drink an Old Coke to remind them of Sun's original formula
> for success.
>
> - from "Sun Reveals A New Dark Side"
From: "Eric Shen" <eric.shen@intel.com>
Hi John,
It's kinda interesting that you mentioned the "New Coke" story.
When I was getting my MBA, the marketing class was taught by an old Pepsi
representative. His take on the "New Coke" rollout and the backlash was
that the Coke executives had everything planned. He noted that once
people started complaining, Coke was very quick in having "Classic" Coke
available. He surmised that it was a can't lose proposition, if the
people accept the New Coke, they win. If they don't, you get a lot of
free advertising and become a hero when the "Classic" coke is rolled out.
- Eric Shen
Intel
---- ---- ---- ---- ---- ---- ----
> Last month I wrote about the customer disgust when Sun required customers
> to sign non-disclosure agreements to get the caches fixed on their broken
> Sun Ultra servers. I received 26 e-mails in response to that column.
>
> - from "Sun's Linux Threat In EDA"
From: David Chapman <dchapman@aimnet.com>
Interestingly, today's (April 5) San Jose Mercury News mentions in brief a
floating point error in the UltraSPARC III CPU as found in "early models of
the Sun Blade 1000" workstation. "No known instances of problems arising
with customers," says the article. Perhaps Sun simply wants to be always
able to brag that it finds problems itself and they don't get out to
customers????
Thanks again for keeping us informed.
- David Chapman
Chapman Consulting Santa Clara, CA
( ESNUG 369 Item 7 ) --------------------------------------------- [04/26/01]
Subject: But Needing 64-bits Kills Any Serious EDA Migration To Linux PCs
> I called Chris and got a copy of his slides. Sure enough, 20 months ago
> NVidia had 10 Linux boxes, 150 Suns, and 100 NT boxes. Last month, NVidia
> had 640 Linux boxes, 270 Suns, and only 70 NT boxes. "We get 2 to 3X
> performance of our Linux boxes over our UNIX workstations and Linux boxes
> are much cheaper, too," Chris said on the phone. "We use Suns only for
> the big jobs because of the current 2 Gig limit in Red Hat."
>
> - from "Sun's Linux Threat In EDA"
From: Jon Stahl <jstahl@avici.com>
Hi John,
I wanted to mention that we are happy users of Linux for several EDA
applications and eagerly anticipate future ubiquitous support from both
software vendors and big iron from hardware vendors.
However, our infrastructure needs to support the "backend" of ASIC projects
that we perform in house. For large devices these applications (e.g.
layout, DRC, nvl, parasitic extraction, xtalk and power droop analysis,
etc.) demand 64 bit OS and hardware, large physical memory images, and
multi-threading with a large number of processors (20+). These are things
that are still either unavailable in the EDA Linux infancy and Intel's
current processor offerings.
So, for the time being an all Linux solution will not suffice.
- Jon Stahl
Avici Systems Massachusetts
---- ---- ---- ---- ---- ---- ----
From: [ The White Rabbit ]
John,
Keep me anon, please
We use PCs and Suns extensively for simulation. Our experience is that the
PCs are faster, but there is a problem: Both NT and Linux have a limitation
on the maximum file size, which limits the size of jobs we can run.
Until version 2.4 of the kernel is released, Linux suffers from a 2 GB
file size limit. This limits the size of simulation jobs we can run under
Linux. NT seems to have a similar problem.
In our benchmarks, NT and Linux seem to give similar performance. I would
really like to see the capacity limitation fixed (which should happen some
time after version 2.4 of the kernel is released)
Even if the PC platforms only give a marginal improvement in speed in
comparison to a Sun (as a previous post suggested), why pay 2x the price
for Sun or HP hardware?
Incidentally, our Cadence AE seemed to have no knowledge of the size
limitation when we were discussing NC-SIM under Linux.
- [ The White Rabbit ]
---- ---- ---- ---- ---- ---- ----
From: Kevin Cameron <dkc@grfx.com>
John,
Sun is not stupid.
I've been running Solaris on PC's since Solaris 2.5, and it may miss some of
the bells & whistles of Linux (which I run, too), but it's darn stable and
I'll trust it over Linux on multiple processors any day. Plus (although I
havn't tried it) you can probably run Linux versions of VCS, etc. through
lxrun, and Solaris is now open source and a free download for X86 -
Links:
http://www.sun.com/intel/
http://groups.yahoo.com/group/solarisonintel/
I talked to an Intel marketing person (not so long ago) who said their
current target to crush was Sun, but was quite umaware that Solaris runs on
X86 (and that would include AMD).
So who is going to give you better support for whatever platform you decide
to use?
Also: You need a 64-bit OS to do LVS and extraction as PC's don't handle
that much data.
- Kevin Cameron
grfx.com
---- ---- ---- ---- ---- ---- ----
From: Moonwook Oh <mwoh@seodu.co.kr>
Dear John,
I've benchmarked and compared some Linux Boxes with Sun's Ultra 60s for
VHDL RTL simulation. Linux Boxes with 1 GHz Pentium III run 2-3 times
faster than Sun's Ultra 60s with small simulation images. (By 'small',
I mean RTL models corresponding to less than 500,000 gates.)
Unfortunately, this 3x speed-up was not the case when the simulation
images become bigger. With most of 'Top-level' models (RTL corresponding
to several hundreds of K gates) Linux Boxes shows ~60% of performance
compared to that of SUN U60s. (It seems that this is because of the
difference in CPU cache and Memory system.)
It was quite a hopeful message for me that many major design groups run
Linux boxes as alternative for Sun machines. So I want to hear from them
about...
1) What is their Linux' performance like with big designs.
2) If they also experiences same performance degradation with big designs.
3) What is the best Linux architecture for EDA jobs (e.g PIII vs Athlon,
BX vs VIA, ....)
Many thanks in advance,
- Moonwook Oh
Seodu InChip, Inc. Seoul, Korea
---- ---- ---- ---- ---- ---- ----
From: "Peter Denyer" <Peter.Denyer@Sun.COM>
John,
Of course, the telling statement is at the end of your column: Linux
solutions are 32-bit and currently limited to 2 gig in Red Hat's varient
of Linux. If you can manage in a 32 bit world with jobs under 2 gig,
the price performance of the "Lintel" boxes is a hard attribute to beat,
but don't believe we're not paying attention to that at Sun.
Based on your own survey at the recent SNUG'01, most of the attendees are
working on designs of over 1M gates, over 200Mhz using .18 micron design
rules. This strikes me that a lot of analysis for designs like this are
not going to fit on a 2 gig Linux box. You also mentioned a significant
requirement for 64bit Design Compiler -- also not fitting on todays Linux
solutions.
You might want to find out how many of your audience see themselves needing
64-bit tools in the next 12 months or so.
- Peter Denyer, EDA Market Development Manager
Sun Microsystems
---- ---- ---- ---- ---- ---- ----
From: "Robert Allgood" <rfre70@email.sps.mot.com>
John,
There aren't a lot of E4500 servers in Motorola SPS - we tend to use
machines with fewer CPUs, since our most difficult jobs require lots of
memory/CPU. There ARE a lot of people around here who bought Ultra60's
when they should have bought Ultra10's for normal user machines and
therefore spent more on a Sun network than they probably should have.
If we ever do make a big push into Linux, I can't see how it will
decrease our Sys Admin load. You still need to get a standardized set
of kernal, graphics, and overlay software which won't get done just by
purchasing Red Hat at CompUSA. You could definitely build a fairly
low-cost Linux farm dedicated to tasks like RIS or other kinds of
front-end regression tests. You could then extend that capability
throughout the design cycle, but I still think most people are keeping
their physical design tasks on traditional Unix for now.
- Robert Allgood
Motorola Semiconductor Products Sector
---- ---- ---- ---- ---- ---- ----
From: Joe Basmaji <jbasmaji@avici.com>
Hi John,
My name in Joe Basmaji, and I work here at Avici with Jon Stahl.
We've been using Linux to run VCS, PrimeTime, and dc_shell. I can tell
you this. I can run my jobs 3x faster on the Linux box. Yes, there may
be limitations at the 2-4 GB mark, but 90% of my compiles/runs don't have
a problem. The problem with the Sun boxes is that not only can we not
keep them up & running (they crash all the time), but their CPU speeds
are too slow, with no end in sight.
- Joe Basmaji
Avici Systems
---- ---- ---- ---- ---- ---- ----
From: "Phil Hoppes" <phoppes@intersil.com>
John,
I read with interest your comments and inputs on the current state of Sun's
and PC/Linux workstations. We came to the same conclusions looking at our
simulation environment. Bottom line, Sun's gave less us than half the EDA
performance at 3x the cost.
On slight differences one might be prudent to keep with Sun.
We put together a 10 by 1 Ghz server farm for our Verilog/C/Matlab
simulations for less than $30K. An equivalent CPU farm in Sun's would
have run over $200K and given us less performance. I can buy a lot of
beans for $170K or in these time, keep more engineers employed. Our
internal IT dept has a love affair with Sun. Granted they are fine
machines but the applications just don't care what hardware they run on.
When the 64-bit applications start to hit in force, Sun should capture
some performance back but hopefully by then Whitney will be shipping
and we can all enjoy Intel's cost curve instead of suffering Sun's
margin curve.
- Phil Hoppes
Intersil BWA
P.S. -- It was interesting too, all this time I thought the big thing
Sun's brought to the party was high speed I/O bandwidth and disk access.
The Dell had FASTER disk I/O access and comparable network speed.
( ESNUG 369 Item 8 ) --------------------------------------------- [04/26/01]
Subject: Yes! We Also Got Mentor To Port Its EDA Tools To Linux, Too!
> In these last two months both Synopsys and Cadence announced they were
> porting many of their more popular EDA tools to Linux.
>
> - from "Sun's Linux Threat In EDA"
From: Michael Dantzer-Sorensen <michael.dantzer@nokia.com>
Hi John,
You wrote: "In these last two months both Synopsys and Cadence announced
they were porting many of their more popular EDA tools to Linux." Mentor
is porting to Linux, too; Design Architect, IC Station, Eldo, ModelSim and
Calibre are all on Linux now.
And you should see this:
From: Michael Dantzer-Sorensen
To: [ Wally Rhines, the CEO of Mentor ]
Dear Mr. Rhines
Remember me? We had a short correspondence around ESNUG 292 #2 and
Linux in June '98.
I just want to thank you personally for doing the Linux 'thing'. :-)
I had the question asked at MUG in '96 but you didn't really want to do
it back then. I'm glad to see that things can change.
Best regards
- Michael Dantzer-Sorensen
Nokia Mobile Phones
From: [ Wally Rhines, the CEO of Mentor ]
To: Michael Dantzer-Sorensen
Michael,
I do remember. Thank you very much; your email actually stimulated
discussion among our management group in 1998 and was a factor in our
(too slow) move to support Linux.
My thanks for your help.
Regards,
- Wally Rhines
President & CEO
Mentor Graphics Corporation
I've been bragging left and right! :)
- Michael Dantzer-Sorensen
Nokia Mobile Phones
---- ---- ---- ---- ---- ---- ----
From: "Marquis Jones" <r40212@email.sps.mot.com>
John,
A couple of years ago, Motorola ported it's version of SPICE to Linux.
It runs great, and it beats having to use and Xterm and an ISDN line
from home or cable modem and a terminal emulator when telecommuting.
Are the Synopsys and Cadence Linux licenses any cheaper? By how much?
- Marquis Jones
Motorola Digital DNA Laboratories
---- ---- ---- ---- ---- ---- ----
From: Michael McIlrath <mbm@vyayama.mit.edu>
John,
You wrote "In these last two months both Synopsys and Cadence announced
they were porting many of their more popular EDA tools to Linux."
Can you point me to the announcements you are referring to? I didn't
find anything more recent than last April at cadence.com.
Thanks!
- Michael McIlrath
MIT
[ Editor's Note: http://www.eetimes.com/story/OEG20000417S0090 - John ]
( ESNUG 369 Item 9 ) --------------------------------------------- [04/26/01]
From: "Don Schuler" <Don_Schuler@avid.com>
Subject: O'Reilly's Linux Clustering Book & Google Uses 6,000 Linux Boxes
John,
We do most of our simulation runs on SUN Ultra-1s, Ultra-2s, E450s.
About 8 months ago we got a Linux box (Dell 700 Mhz with Redhat 6.1.1) and
did some simulation tests use Modelsim/VHDL. When all the data was local
on the Linux box, results where great. Performance was equal to or greater
then what one would expect from the clock rate differences between the Sun
and Dell. So price/performance wise it looked like a big win for the Linux
box.
However, when we ran cases where the data was distributed around the network
(LAN), with our work library on one system, ASIC/FPGA libraries on another
system, logfiles going to a 3rd system, the simulations on the SUNs ran much
faster (3-6x). Since multiple people work on a project and we use LSF to
distribute runs to a server farm, we are always using distributed data.
Anyway what are we doing wrong? How are people making Linux simulation
farms get better performance? I heard that NSF on Redhat 6.2 is slow, but
could that be the full story? Are people using Linux for file servers as
well as compute servers?
- Don Schuler
Avid Technology, Inc.
---- ---- ---- ---- ---- ---- ----
From: Richard_J_McWaters@raytheon.com
Another interesting piece of information on Linux. Google (www.google.com)
is a very efficient web search tool. Try it, you will be impressed. They
use Linux to create massive clusters. When I say massive, they were using
over 6,000 Intel boxes last summer and expected to have over 10,000 by the
end of the year (2000).
Their system guru is a former med student that got interested in search
engines and figured out a way to do the clustering. With the way the system
is arranged, it allows for failures of individual units without major effect
on the system. The redundancy is staggering and the performance of the
system (from my end user standpoint) is impressive.
I pay a lot of money in hardware and support, but since I have a single
server, I have a single point of failure. In fact, classic failure analysis
on my system is dismal. Google has a system that is cheap, fast and
reliable. There is an O'Reilly book on Linux clustering, so it should not
be too hard to get this set up. Unfortunately, the IT organization at my
company (as well as most other companies) can't see the forest for the
trees. We don't even have anyone looking into the potential of Linux!
I think you have hit on a very important area for the future of design
engineers. It doesn't matter if you are using Synopsys, Mentor or ProE
(mechanical design), the need for faster, cheaper, more reliable computing
resources continues to grow. With the moves toward a common desktop and
with companies like MainSoft (who port PC applications to UNIX), the
potential to move to Linux as a common desktop is real. I get Sun news
all the time, and they have turned a blind eye to this phenomenon. They
could be sitting still while the world passes them by.
Finally, some simple financials. It is a lot easier to get a few thousand
to incrementally upgrade the computer system than get $75K to buy one
server. And talk about depreciation! My 3 year old 3500 that cost me over
$100K is now worth about $1.50.
- Richard McWaters
Raytheon
( ESNUG 369 Item 10 ) -------------------------------------------- [04/26/01]
From: Mark Wroblewski <markwrob@colorado.cirrus.com>
Subject: Hmmm... Synopsys Ported Everything *But* EPIC To Linux... Hmmm...
Hi John,
In case you haven't seen this, here's a listing I made last month from SNPS'
ftp site. So they have already got their main tools released for Linux.
ftp> dir */*linux*
200 PORT command successful.
150 Opening ASCII mode data connection for /bin/ls.
12221 Feb 23 20:23 dwdu_v2000.11/dcl_2000.11_linux.PDF
316132 Feb 23 20:23 dwdu_v2000.11/dwdu_2000.11_linux.tar.Z
45755992 Feb 13 09:12 mp_v2000.11-7/mempro_x86_linux.zip
84820 Feb 13 09:12 mp_v2000.11-7/sl_unzip.x86_linux
147755832 Mar 26 22:53 pt_v2000.11-SP1/pt_2000.11-SP1_linux.tar.Z
760405 Nov 29 08:35 scl_v1.1a-1/scl_1.1a-1_linux.tar.Z
1590647 Oct 25 19:28 scl_v1.1a/scl_1.1a_linux.tar.Z
84820 Mar 15 06:09 sm_v2000.11-11/sl_unzip_x86_linux.exe
369485898 Mar 15 06:56 sm_v2000.11-11/smartmodel_x86_linux.zip
245889225 Mar 26 19:56 syn_v2000.11-SP1/syn_2000.11-SP1_linux.tar.Z
12221 Mar 6 00:49 syn_v2000.11/dcl_2000.11_linux.PDF
167065791 Mar 6 00:49 syn_v2000.11/syn_2000.11_linux.tar.Z
17291727 Jun 16 2000 vcs_v5.2/intel_i686_linux_2.2.tar.gz
17299045 Jun 16 2000 vcsi_v5.2/intel_i686_linux_2.2.tar.gz
26792640 Nov 17 02:08 vera_v4.5/vera-4.5-linux2.2.5.tar.gz
5537127 Oct 20 18:29 vs_v4.0.2/virsim_4_0_2_p2_linux.tgz
226 Transfer complete.
Used to be, Synopsys gave lip service (like most other EDA vendors) to Linux
by putting one tool (maybe FPGA Express?) on Linux. But with synthesis and
PrimeTime on Linux, looks like they're committed. No Epic tools though.
Perhaps that indicates their plans to phase Epic functions into their other
tools and eventually get rid of them entirely.
It would be interesting to know how many users are running these current
Synopsys tools on Linux boxes, and whether they've experienced any more than
the usual number of issues....
- Mark Wroblewski
Cirrus Logic Broomfield, CO
( ESNUG 369 Item 11 ) -------------------------------------------- [04/26/01]
Subject: Intel?, Athlon?, Abit?, Asus? -- What Is In NVidia's Linux Farm?
> I called Chris and got a copy of his slides. Sure enough, 20 months ago
> NVidia had 10 Linux boxes, 150 Suns, and 100 NT boxes. Last month, NVidia
> had 640 Linux boxes, 270 Suns, and only 70 NT boxes.
>
> - from "Sun's Linux Threat In EDA"
From: John Jakson <johnjakson@earthlink.net>
Hi, John,
I would be very interested to know a bit about the configuration of those
640 nVidia PC boxes, especially the motherboard & Intel v Athlon. I
suspect Abit, Asus, Gigabyte are used. Are they using fancy high end
SCSI drives? It could help me choose more wisely my next configs
components. My 1G Athlon sucks, freezes after a few mins or a few hours.
What can you expect from W2k & a no name board?
It's a shame Athlons are not yet multi processor. It should be very
interesting to see the Hammerhead SMP 64-bit Athlon, will it nail both
Sun & Intel? I don't see EDA going Itanium, & 32bit x86 is the only
thing saving Sun's rear end.
- John Jakson
( ESNUG 369 Item 12 ) -------------------------------------------- [04/26/01]
From: Simon Read <simon@formasic.com>
Subject: Simon Read -- Linux On SPARCs & Beowulf Linux Network Clusters
John,
Two things you didn't mention in your article that are well worth
knowing:
Linux is available for SPARC and most of the distributions support it
(RedHat and Debian for sure). I use it on my SPARC 20 at home. Some
applications that don't make use of Solaris specific features (kernel
supported threads for example) will run in binary emulation using
provided libraries. As a former EDA developer, I suspect that most
EDA applications would run in emulation, because they are designed to
be portable to HP-UX, AIX, NT and so forth.
Linux offers another advantage: Beowulf. Beowulf clusters are groups
of stripped down Linux boxes with special networking software. This
allows them to be harnessed as an efficient multi-processor without
special networking hardware. If EDA vendors were to appropriately
modify their applications they could make very effective use of these
scalable systems. In fact, I don't believe they can keep up by relying
on fast uniprocessors, but that's a whole other story :)
- Simon Read
Form ASIC
============================================================================
Trying to figure out a Synopsys bug? Want to hear how 11,000+ other users
dealt with it? Then join the E-Mail Synopsys Users Group (ESNUG)!
!!! "It's not a BUG, jcooley@world.std.com
/o o\ / it's a FEATURE!" (508) 429-4357
( > )
\ - / - John Cooley, EDA & ASIC Design Consultant in Synopsys,
_] [_ Verilog, VHDL and numerous Design Methodologies.
Holliston Poor Farm, P.O. Box 6222, Holliston, MA 01746-6222
Legal Disclaimer: "As always, anything said here is only opinion."
The complete, searchable ESNUG Archive Site is at http://www.DeepChip.com
|
|
