Purchase of Large UNIX Server, seeking info!

Purchase of Large UNIX Server, seeking info!

Post by System Manag » Fri, 03 May 1996 04:00:00



The Computer Science Department at the University of Nebraska-Lincoln
is in the process of looking at replacing our central login server.
Every student taking Comp Sci classes is given an account on this Server.
This is therefore our central E-mail, programming, www & research machine.
I would really appreciate some feedback from sites that have a similar
computing environment as we do.  We would like to be spared the regrets you
have, the vendor sales pitches you fell for, the support and portability
nightmare you still experience. I'm going to lay out our computing
environment, followed by some issues that I'm concerned with.  Feel free to
address any subset of these issues and others that I have not had the
foresight to be aware of yet!  Use any format you please. You can send me

Computing Environment:
----------------------

Hardware Environment:
        Server : 1 Central Login Server and a few secondary servers.
        Clients: - ~75 Xterminals.
                 - ~50 Workstations.
                 - Terminal Servers scattered across campus.

Usage Characteristics:
                                Present         Potential
                                -------         ---------
        Accounts            :   1000            1500
        Simultaneous Logins :   100             200
        Number of Processes :   600             2000+

        Usage Portfolio     :   About 1/4 of our processes are X applications
                                like xrn, netscape, xterm, xdbx, xfig,
                                     ghostview.
        General Usage       :   E-mail, Network News, WWW, Programming,
                                TeX/Latex.

Our Central login server is currently an 4cpu, SGI 4D/440, with 196MB RAM.
we have had to limit the number of simultaneous processes to 600. After the
number of simultaneous processes hits the high 500s the machine goes
from being usable to being barely so. The main bottlenecks we feel are
are memory and I/O. I/O waits are up in the high 30% and swap I/O waits are
up in the high *s as the processes count creeps up into the high 500s.

Here are some questions I have for sites like ours:
---------------------------------------------------

        - The hardware characteristics of your central login server.

        - Very brief summary of your environment and usage characteristics
          to give me an image of your usage portfolio.

        - Are you happy with the Interactive responsive time?

        - How does the interactive response time hold up when
          someone fires 20 cpu intensive tasks?

        - If you feel you have bottle necks on your system what
          are they?

        - How has the machine worked out? Nice things, bad things.

        - Are you happy with administration and Sys monitoring tools?
          Want to tell me a little about these tools?

        - Do you have any pointers/stories to help us along.
                - Support nightmares?
                - Application portability stories?
                - User code nightmares?
                - Sysadmin peculiarities?
                - Anything else?

        - If you run a 64bit OS:
                - Is it truly backward compatible with its 32bit predecessor?
                - what are the problems you have encountered?
                - Do you find that the XMB of memory you have does not
                  go as far, with pointers and integers now occupying
                  64bits instead of 32bits. How noticeable is this?
                - Anything else about your 64bit OS?

        - Things you would have done differently. Changes you would
          have made in your configuration if you could do it all over again.

-------------------------------------------------------------------
Charles Daniel

Computer Science and Engineering                    115 Ferg (0115)
University of Nebraska-Lincoln                        (402)472-9477
-------------------------------------------------------------------

 
 
 

Purchase of Large UNIX Server, seeking info!

Post by Doug Siebe » Fri, 03 May 1996 04:00:00



>Usage Characteristics:
>                                Present         Potential
>                                -------         ---------
>        Accounts            :   1000            1500
>        Simultaneous Logins :   100             200
>        Number of Processes :   600             2000+

>        Usage Portfolio     :   About 1/4 of our processes are X applications
>                                like xrn, netscape, xterm, xdbx, xfig,
>                                     ghostview.
>        General Usage       :   E-mail, Network News, WWW, Programming,
>                                TeX/Latex.

>Our Central login server is currently an 4cpu, SGI 4D/440, with 196MB RAM.
>we have had to limit the number of simultaneous processes to 600. After the
>number of simultaneous processes hits the high 500s the machine goes
>from being usable to being barely so. The main bottlenecks we feel are
>are memory and I/O. I/O waits are up in the high 30% and swap I/O waits are
>up in the high *s as the processes count creeps up into the high 500s.

Only 196MB of RAM for that many users?!  There's your problem right there.
Whatever you buy (you will have people recommending for/against every major
vendor's machines, probably in roughly equal numbers to the sales of each
vendor) you should get 512M - 1G of RAM, minimum.  If you are having disk
I/O bottlenecks on an interactive system it is mainly proof that you don't
have enough memory -- your buffer cache is too small.  Its much easier and
cheaper to beef up your RAM and have a bigger buffer cache than it is to
purchase a faster I/O system and arrays of striped disks to take advantage of
the performance improvements available.  You could probably double the RAM in
your current box and get 50% more capacity today if that's what you wanted to
do (though that thing is pretty old so I can see why you want to replace it)

--
Doug Siebert              || "Usenet is essentially Letters to the Editor
University of Iowa        ||  without the editor.  Editors don't appreciate

(c) 1996 Doug Siebert.  Redistribution via the Microsoft Network is prohibited.

 
 
 

Purchase of Large UNIX Server, seeking info!

Post by Ron Natal » Fri, 03 May 1996 04:00:00


We never used to put that many users on a single machine if all they
are doing is doc processing and other lightwieght stuff.  Grab four
or five largish servers and divide your load between them.  Cross
mount the disks accross them.

-Ron

 
 
 

Purchase of Large UNIX Server, seeking info!

Post by Eric Chandl » Thu, 09 May 1996 04:00:00


Three suggestions (mutually exclusive)

Suggestion 1:
SparcServer 2000
Rather expensive, but would handle the load and the eventual growth.  If you
got one of these, you would probably need about 8 CPUs and 1 Gig of ram.  As
far as disk space, college students use alot.  Get a Sparc Storage Array of
AT LEAST 32 gig.  Usually these things come in a big box which is expandable.
The OS would most likely by Solaris 2.5.  Naming services could be debated.

Cost: (eek!) Could be as high as $1,000,000.  It WOULD handle your situation

Suggestion 2:
Spread out the load on a bunch of smaller servers.  Used SparcServer 670s and
690s could be probably picked up for as little as $50,000.  They are deemed
"worthless" to many large firms (like mine), since they usually only handle
2-4 cpus and 256 meg.  If you put maybe 500-1000 users per machine, you can
even out the load a bit.  For each server, you would probably need 20 gig of
disk space.  

Suggestion 3:
Add more workstations to your existing system and disallow direct logins to
the server.  Rather, force them to log into the workstations, even from
remote sites.  This way you can pound the workstations to death with the
x-windows stuff, while the server supplies the e-mail, NFS, NEWS, and the
"brains" of the network.

Since I am a Sun Solaris and IBM SP/2 administrator, I can only give you the
Sun solution to the problem.  These are by no means the ONLY options you have.
You have probably 500 machines out there that could do the job.  Some have
better support for the customers than others, though.

Eric Chandler

 
 
 

Purchase of Large UNIX Server, seeking info!

Post by Rob Raim » Thu, 09 May 1996 04:00:00


I am a sys admin of an SGI network of > 70 users. (Mostly developers)
We automount home accounts and use clearcase (rev. control software)
Our server used to be a 4D/440S, we replaced it with a Challenge L, 4/200mHz cpu's,
512 megs ram, 1 IO/4 w/mez. We also purchased an SGI raid5 with 2 SP's so we
could stripe across 2 controllers.
First the server: The challenge L has plenty of CPU power and never has a shortage
of RAM. We were having a problem with I/O max'ing out and setting buffers to the
max didn't do it. I will be adding a second IO4 and another array and all
should be well. The combination of home accounts and clearcase makes for a
very heavy I/O load.
The SGI RAID5 has not been going well. I have had 2 drives fail in 4 weeks, the
set-up software, after 2 revisions, 2 flash code changes and a number of patches
still doesn't work right. Trying to use the GUI based set-up causes a seg fault
and a core dump every time. The command line is not easy to use. So far I have
plugged a monitor into the raid and made changes there.  sigh...  To be fair
I was told that I had the first 2 SP RAID5 ( for striping) that SGI had sold.

Do I like the server? Yes, it is fast and very dependable. If you have lots of users,
and it sounds like you do, I suggest that you stripe across 2 or more scsi
 controllers.

: The Computer Science Department at the University of Nebraska-Lincoln
: is in the process of looking at replacing our central login server.
: Every student taking Comp Sci classes is given an account on this Server.
....snip.....

:  
: Computing Environment:
: ----------------------

--

     Any sufficiently advanced magic -
                    is indistinguishable from technology


    Cineon Systems-TAC      Pager: (800) skypage #227-2807                                        
    Eastman Kodak Company   Phone: (716) 253-9330            
    901 Elmgrove Rd.                    KMX:         233-9330            
    2/4 EP                              FAX:   (716) 253-9467            
    Rochester, NY USA 14653-5301
   <----------------------------------------------------------->      

 
 
 

Purchase of Large UNIX Server, seeking info!

Post by bill davids » Thu, 09 May 1996 04:00:00



|
|
| The Computer Science Department at the University of Nebraska-Lincoln
| is in the process of looking at replacing our central login server.
| Every student taking Comp Sci classes is given an account on this Server.

That starts out sounding like a not great idea.

| This is therefore our central E-mail, programming, www & research machine.

That sounds like a really terible idea!

Having run Internet servers at GE's corporate R&D Center, Prodigy,
and about a half dozen other sites including my own, I strongly
suggest running your web site and mail server on separate
machine(s), to keep the load fairly constant.

Web servers are also subject to load fluctuations, but if you limit
the number of processes you can keep the load from becoming a
problem if a student puts up something of wide interest. RPI had a
problem like this and resolved it through good system administration
(totally without my help, I just know about it).

Mail is generally a critical organization resource, RAID5 with hot
backup and a backup server are very good ideas. You should decide
how long you can afford to have mail down and plan for preventing
longer outages. Honestly for your load level a small mail machine
like a Pentium running Linux would probably do just fine, and could
do DNS as well.

You can get large servers from lots of places. Sun has the advantage
of being the first or second port for virtually any software, and
IBM has the advantage of pretty good stuff and total financial
stability. HP is going to change CPUs soon, doing the deal with
Intel, that may make existing machine obsolete, or affect their
upgrade availability. I view SGI as more aimed at graphics, but they
make some nice servers (which I haven't used enough to evaluate).

Both Sun and IBM offer good MP servers, Sun the 2000 and IBM the
R24/R30/J30 class machines. IBM offers the SP02 cluster, which
offers the expansion of fast nodes, very fast node, or really fast
SMP nodes, all connected with a multi-megabit data bus and some very
nice redundancy features and remote system management features.

Since all companies offer educational discounts, I think you have to
pick a good candidate and then haggle. I like the SP02, since you
can run a low cost node as a web server, one as a mail server, etc,
and still press a development node into service in case of a
failure. And the ability to add processors to a node or nodes to a
cluster gives you a lot of scaling if you need it.

Hope this is useful!
--

"The secret to procrastination is to put things off until the
 last possible moment - but no longer"  -me

 
 
 

Purchase of Large UNIX Server, seeking info!

Post by Peter Clinc » Thu, 09 May 1996 04:00:00



> Three suggestions (mutually exclusive)

> Suggestion 1:
> SparcServer 2000
> Rather expensive, but would handle the load and the eventual growth.  If you
> got one of these, you would probably need about 8 CPUs and 1 Gig of ram.  As
> far as disk space, college students use alot.  Get a Sparc Storage Array of
> AT LEAST 32 gig.  Usually these things come in a big box which is expandable.
> The OS would most likely by Solaris 2.5.  Naming services could be debated.

> Cost: (eek!) Could be as high as $1,000,000.  It WOULD handle your situation

Modified suggestion 1:
The new UltraSPARC based servers have just been launched (check out www.sun.com
for the details).  There's quite a range to choose from, and the available models
are faster and more flexible for configuartion than the old SC2000s.  Faster
interconnects as well as faster processors, plus hot plug in facility and new
system monitoring so it should be more reliable too.  Most should ship in volume
from the end of this month, so probably a better bet than the old 2000s

As for college students using a lot of space, they actually use more than that!
Our sysadmin was asked an interesting disk space question recently, along the
lines of...
"Can you modify vi for me, so when I want to save it checks to see if there's
enough space, and if not, it deletes unimportant files until there's enough?"
Carol enqiuires on a definition of "unimportant", and it transpires this means
"does not belong to me"!  She is very restrained, and does not use the word
"pratt" in her reply...

Pete.
--
Peter Clinch                            Dundee University & Teaching Hospitals
Tel 44 1382 660111 extension 3637       Medical Physics, Ninewells Hospital
Fax 44 1382 640177                      Dundee DD1 9SY Scotland UK

 
 
 

Purchase of Large UNIX Server, seeking info!

Post by Butch De » Fri, 10 May 1996 04:00:00




> |> Three suggestions (mutually exclusive)
> |>
> |> Suggestion 1:
> |> SparcServer 2000
> |> Rather expensive, but would handle the load and the eventual growth.  If you
> |> got one of these, you would probably need about 8 CPUs and 1 Gig of ram.  As
> |> far as disk space, college students use alot.  Get a Sparc Storage Array of
> |> AT LEAST 32 gig.  Usually these things come in a big box which is expandable.
> |> The OS would most likely by Solaris 2.5.  Naming services could be debated.
> |>
> |> Cost: (eek!) Could be as high as $1,000,000.  It WOULD handle your situation
> |>
> |>
> |> Suggestion 2:
> |> Spread out the load on a bunch of smaller servers.  Used SparcServer 670s and
> |> 690s could be probably picked up for as little as $50,000.  They are deemed
> |> "worthless" to many large firms (like mine), since they usually only handle
> |> 2-4 cpus and 256 meg.  If you put maybe 500-1000 users per machine, you can
> |> even out the load a bit.  For each server, you would probably need 20 gig of
> |> disk space.  
> |>

> I'd certainly think hard about buying a 2000 these days. The ultras blow
> these (and 1000 class machines) pretty much away in all regards and cheaper
> for the most part. If you really need a huge enterprise server, I'd go
> with sun's new line of servers and skip the 2000 all together. I'd also
> skip the 670s and 690s too. A 170E will make short work of them on just
> about everything, except if you have jobs that take advantage of
> parallelization (multi-threaded jobs), then maybe these machines will perform
> better.

> Unfortunately, I didn't see the original question, but, the Ultras are
> very nice machines. the lm_bench benchmarks by Larry McVoy put the Ultras
> WAY above a 1000. It's not even funny how far apart they are in every
> category. In fact, the 1000 places last in most tests, even behind Pentium
> systems running Linux. However, these are all single CPU type benchmarks,
> so the 1000 with 4 CPUs may do well on multi-threaded applications when
> compared to some of the other machines. But for the price of an Ultra,
> you may be able to get several of them for a multi-CPU 1000.

> Look carefully at your options, and strongly consider the newer technology
> (ultra, enterprise server) over the older (2000, 2000, 690, 670).

You probably need to look at the 3000 or 4000 from sun.  They are much cheaper
than a 2000 and more expandable than an Ultra1 or Ultra2.
take a look on suns web page for more info (http://www.sun.com/) on them.

--
#include <std/*>                           '91 FLT
The Butcher                  

--------------------------------------------------------------------------------

 
 
 

Purchase of Large UNIX Server, seeking info!

Post by Max Waterma » Sat, 11 May 1996 04:00:00



> I am a sys admin of an SGI network of > 70 users. (Mostly developers)
> We automount home accounts and use clearcase (rev. control software)
> Our server used to be a 4D/440S, we replaced it with a Challenge L, 4/200mHz cpu's,
> 512 megs ram, 1 IO/4 w/mez. We also purchased an SGI raid5 with 2 SP's so we
> could stripe across 2 controllers.
> First the server: The challenge L has plenty of CPU power and never has a shortage
> of RAM. We were having a problem with I/O max'ing out and setting buffers to the
> max didn't do it. I will be adding a second IO4 and another array and all
> should be well. The combination of home accounts and clearcase makes for a
> very heavy I/O load.

I can second this. ClearCase is extremely heavy on IO since it has to maintain integrety
of it versioned object (data)base (VOB). Also, each user (of a view) has heavy use of IO
for the versions of s/w they have checked out. The release notes advise increasing the
disk cache limits substantially. We had a satisfactory setup for 4 developers using a
Challenge M (I^2) R4400/150 with two 1GByte drives striped across it's two controllers.
Rob (the author of the post I am replying to) sounds like he has a similar setup but
with a much higer load.

All I am saying is that don't underestimate the load that he puts his h/w under - it is
substantial - sounds like his h/w works satisfactorily too - appart from the RAID.

NB I thought you could stripe satisfactorily on the Challenge L's diff. SCSI bus.
Doesn't this give you the same benefits (ie IO speedup)?

Max.

> The SGI RAID5 has not been going well. I have had 2 drives fail in 4 weeks, the
> set-up software, after 2 revisions, 2 flash code changes and a number of patches
> still doesn't work right. Trying to use the GUI based set-up causes a seg fault
> and a core dump every time. The command line is not easy to use. So far I have
> plugged a monitor into the raid and made changes there.  sigh...  To be fair
> I was told that I had the first 2 SP RAID5 ( for striping) that SGI had sold.

> Do I like the server? Yes, it is fast and very dependable. If you have lots of users,
> and it sounds like you do, I suggest that you stripe across 2 or more scsi
>  controllers.

> : The Computer Science Department at the University of Nebraska-Lincoln
> : is in the process of looking at replacing our central login server.
> : Every student taking Comp Sci classes is given an account on this Server.
> ....snip.....

> :
> : Computing Environment:
> : ----------------------

> --

>      Any sufficiently advanced magic -
>                     is indistinguishable from technology


>     Cineon Systems-TAC      Pager: (800) skypage #227-2807
>     Eastman Kodak Company   Phone: (716) 253-9330
>     901 Elmgrove Rd.                    KMX:         233-9330
>     2/4 EP                              FAX:   (716) 253-9467
>     Rochester, NY USA 14653-5301
>    <----------------------------------------------------------->

--

  /  /  /  _      /   /  _ __ __  _  __   _
 /  /  / /_/ |/  / / / /_/ / /_ /_/ /// /_/ /| /
/  /  / / / /|  /_/_/ / / / /_ /  |/// / / / |/
 
 
 

Purchase of Large UNIX Server, seeking info!

Post by Martin Hargreav » Mon, 13 May 1996 04:00:00



>I am a sys admin of an SGI network of > 70 users. (Mostly developers)
>We automount home accounts and use clearcase (rev. control software)
>Our server used to be a 4D/440S, we replaced it with a Challenge L, 4/200mHz cpu's,
>512 megs ram, 1 IO/4 w/mez. We also purchased an SGI raid5 with 2 SP's so we
>could stripe across 2 controllers.
>First the server: The challenge L has plenty of CPU power and never has a shortage
>of RAM. We were having a problem with I/O max'ing out and setting buffers to the
>max didn't do it. I will be adding a second IO4 and another array and all
>should be well. The combination of home accounts and clearcase makes for a
>very heavy I/O load.

Since the original poster already has SGI and it seems to work for
them a Challenge (L or XL) sounds about right. The Challenge machines
are excessively good at I/O, esp. the big ones (XL's backplanes shift
about 1.3Gb/s to I/O subsystems AFAIR)

I used to work at a place where we used SGI, a 4D/480 w 256Mb, a
2xR4400 Challenge with 2Gb, and a 4 CPU PowerChallenge. IRIX 5.3 seems
to scale nicely to many users from what I've heard and the machines
were adequate for our applications (mostly computational chemistry -
CPU, memory, and disk intensive but only about ten users)

Quote:>Do I like the server? Yes, it is fast and very dependable. If you have lots of users,
>and it sounds like you do, I suggest that you stripe across 2 or more scsi
> controllers.

Goo advice.

Quote:>     Any sufficiently advanced magic -
>                    is indistinguishable from technology

      Any technology distinguishable from magic -
                     is insufficiently advanced

##################################################################

# Director, Datamodel Ltd                                Chemist #
# Contract Unix system admin/Unix security              Sysadmin #  
##################################################################

 
 
 

Purchase of Large UNIX Server, seeking info!

Post by George Herbe » Tue, 14 May 1996 04:00:00



Quote:>Modified suggestion 1:
>The new UltraSPARC based servers have just been launched (check out
>www.sun.com for the details).  There's quite a range to choose from, and
>the available models are faster and more flexible for configuartion than
>the old SC2000s.  Faster interconnects as well as faster processors,
>plus hot plug in facility and new system monitoring so it should be more
>reliable too.  Most should ship in volume from the end of this month,
>so probably a better bet than the old 2000s

I think the "most should ship in volume from the end of this month" is
optimistic.  The Ultra 2 isn't even shipping in volume yet...

However, yes, the Ultra servers should be significantly better than
the SparcCenter / SparcServer machines when they ship.

-george william herbert

 
 
 

Purchase of Large UNIX Server, seeking info!

Post by Stephen Mulli » Thu, 16 May 1996 04:00:00


I've not seen the original spec for what your doing, sounds big..but
with an EDU discount I'd go for Digital's new Alpha 4000 box, Rawhide.
The Alphas beat all the competition, check one out.

Storage and such, design it with assistance from vendors, _then_ price
each.  Remember to get the battery for your raid units so you can enable
write cache on the RAID controller. (beware of rigged benchmarks via
this most important feature)

Yup...ClearCase is a bear!
--
Stephen Mullin       Digital Unix O/S & Cluster Support & Services

 
 
 

Purchase of Large UNIX Server, seeking info!

Post by Kenneth Scott Beth » Thu, 16 May 1996 04:00:00


:
:
: The Computer Science Department at the University of Nebraska-Lincoln
: is in the process of looking at replacing our central login server.
<SNIP>
:  
: Usage Characteristics:
:                                 Present         Potential
:                                 -------         ---------
:         Accounts            :   1000            1500
:         Simultaneous Logins :   100             200
:         Number of Processes :   600             2000+
:  
:         Usage Portfolio     :   About 1/4 of our processes are X applications
:                                 like xrn, netscape, xterm, xdbx, xfig,
:                                      ghostview.
:         General Usage       :   E-mail, Network News, WWW, Programming,
:                                 TeX/Latex.
:  

Man I tell you what, I can't unbderstand why people have this idea that you MUST run
everything on one big overpriced underpowered server.  Your "Usage Characteristics"
above look like what any ISP would have to face..  They solve the "problem" in many
different ways, but the *BEST* I think is to use more than one server, use some sort of
load-balancing, and split up the work on as many servers as it will take.  Terminal
Servers take the load off a cpu, and you say you will be using them, great, plan your
whole network like this, have two or three www servers, at least a couple Nameservers,
mailservers, database servers, news servers, etc, etc.  Don;t listen to the hype about
Ultra's, the OS isnt there to support the hardware yet, stick to the basics that we all
know work well..  SS10's, 20's, and yes 690MP's  These are dirt cheap, and with enough
of them spread around you will have a nice system that will easily take the kind of
abuse your power-series SGI is taking now...  

Now for future options, when old servers die or become less able to deal with the
stress, downgrade them to easy tasks, like once that 4 processor SS20 that handles 1/3rd
of your web server needs gets a little old, downgrade it to a mail-serve.  maybe then
the Ultra's and such will be up to par with a nice 64Bit (*compatible*) OS.

-Scott

--
=    K e n n e t h   S   B e t h k e   =
-----------------------------------------------------------------------
GSTek Corporation                   We BUY/SELL and TRADE SUN Computers

-----------------------------------------------------------------------
<a href="http://www.ezy.net/gstek/">  HTTP://WWW.EZY.NET/GSTEK/   </a>

 
 
 

Purchase of Large UNIX Server, seeking info!

Post by Rob You » Fri, 17 May 1996 04:00:00





>:
>:
>: The Computer Science Department at the University of Nebraska-Lincoln
>: is in the process of looking at replacing our central login server.
>:
>: Usage Characteristics:
>:                                 Present         Potential
>:                                 -------         ---------
>:         Accounts            :   1000            1500
>:         Simultaneous Logins :   100             200
>:         Number of Processes :   600             2000+
>:  
>:         Usage Portfolio     :   About 1/4 of our processes are X applications
>:                                 like xrn, netscape, xterm, xdbx, xfig,
>:                                      ghostview.
>:         General Usage       :   E-mail, Network News, WWW, Programming,
>:                                 TeX/Latex.
>:  

>Man I tell you what, I can't unbderstand why people have this idea that you
>MUST run everything on one big overpriced underpowered server.  
>Your "Usage Characteristics" above look like what any ISP would have to
>face..  They solve the "problem" in many different ways, but the *BEST* I
>think is to use more than one server, use some sort of load-balancing, and
>split up the work on as many servers as it will take.  Terminal Servers take
>the load off a cpu, and you say you will be using them, great, plan your
>whole network like this, have two or three www servers, at least a couple
>Nameservers, mailservers, database servers, news servers, etc, etc.  Don;t
>listen to the hype about Ultra's, the OS isnt there to support the hardware
>yet, stick to the basics that we all know work well..  SS10's, 20's, and yes
>690MP's  These are dirt cheap, and with enough of them spread around you will
>have a nice system that will easily take the kind of abuse your power-series
>SGI is taking now...  

        Perhaps you wish to look at what Lycos has done and what others
        are doing.  Lycos has dumped their distributed model and gone
        with 2 powerful "main-frame killer" UNIX servers.  The distributed
        model has its weaknesses:

                1) Network traffic.
                2) Adequate memory on each distributed machine ($$$)  **
                3) Software distribution and maintenance
                4) System maintenance (many more components)

        Another difficulty is the people cost involved, this could add up to
        more money than the hardware as you have to increase the head count
        to maintain more machines.  Definitley spend more time, so you either
        overwork the existing staff or hire more people.  Which will it be???

        What you wish to do is purchase a server you **KNOW** will scale for
        your modest requirements.  A certain OEM of mainframe killers
        out of Maynard, Mass. comes to mind.  They have managed to index all of
        the available Internet nicely, a good indicator of scalability.

        Stick with the centralized approach.

                                Rob

** Look at this closely.  How much for each machine?

 
 
 

Purchase of Large UNIX Server, seeking info!

Post by Butch De » Fri, 17 May 1996 04:00:00




> :
> :
> : The Computer Science Department at the University of Nebraska-Lincoln
> : is in the process of looking at replacing our central login server.
> <SNIP>
> :  
> : Usage Characteristics:
> :                                 Present         Potential
> :                                 -------         ---------
> :         Accounts            :   1000            1500
> :         Simultaneous Logins :   100             200
> :         Number of Processes :   600             2000+
> :  
> :         Usage Portfolio     :   About 1/4 of our processes are X applications
> :                                 like xrn, netscape, xterm, xdbx, xfig,
> :                                      ghostview.
> :         General Usage       :   E-mail, Network News, WWW, Programming,
> :                                 TeX/Latex.
> :  

> Man I tell you what, I can't unbderstand why people have this idea that you MUST run
> everything on one big overpriced underpowered server.  Your "Usage Characteristics"
> above look like what any ISP would have to face..  They solve the "problem" in many
> different ways, but the *BEST* I think is to use more than one server, use some sort of
> load-balancing, and split up the work on as many servers as it will take.  Terminal
> Servers take the load off a cpu, and you say you will be using them, great, plan your
> whole network like this, have two or three www servers, at least a couple Nameservers,
> mailservers, database servers, news servers, etc, etc.  Don;t listen to the hype about
> Ultra's, the OS isnt there to support the hardware yet, stick to the basics that we all
> know work well..  SS10's, 20's, and yes 690MP's  These are dirt cheap, and with enough

have you tested any Ultra systems?  were are you getting the idea that the OS is
"not there"?  I have run and tested the ULTRA systems and the OS is the same OS
(Solaris 2.5.1) that runs on the SS10,SS20, and SS690MP.  The SS10 and SS690MP are
no longer sold by Sun.  The SS690MP my not be supported in Solaris 2.6 either.
From my tests the $9k Ultra 1/140 was far faster than the SS10 or SS690MP in any
configuration they support.  We did not have a SS20/812 to test against the Ultra1/140,
it would have probably been just slightly faster than the Ultra 1/140 but
the Ultra 1/140 comes in around $9k and the SS20/812 comes in at over $25k
For more info on our comparison of the systems check out
http://www.ait.nrl.navy.mil/modsaf/modsaf.html

Quote:> of them spread around you will have a nice system that will easily take the kind of
> abuse your power-series SGI is taking now...  

> Now for future options, when old servers die or become less able to deal with the
> stress, downgrade them to easy tasks, like once that 4 processor SS20 that handles 1/3rd
> of your web server needs gets a little old, downgrade it to a mail-serve.  maybe then
> the Ultra's and such will be up to par with a nice 64Bit (*compatible*) OS.

Two things here.  mail servers for large sites tend to take more of a beating than
the web servers.  second, why does the OS have to be 64Bit "(*compatible*)" to use
the 64bit machine?  Since you are recomending useing 32Bit machines why would the
Ultra need 64bits?  You might also whant to check some things Solaris 2.5.1 does
give you access to some 64bit options.  

Quote:

> -Scott

> --
> =    K e n n e t h   S   B e t h k e   =
> -----------------------------------------------------------------------
> GSTek Corporation              We BUY/SELL and TRADE SUN Computers

apparently you don't "BUY/SELL or TRADE" many Ultra Sparc Computers


> -----------------------------------------------------------------------
> <a href="http://www.ezy.net/gstek/">  HTTP://WWW.EZY.NET/GSTEK/   </a>

--
#include <std/*>                           '91 FLT
The Butcher                  

--------------------------------------------------------------------------------
 
 
 

1. Help with LARGE server purchase



Some of what your doing is unclear, but assuming you don't absolutely
need one large server here is my suggestion:

An NFS network of Linux 486 PC's each handling X out of the 200 users. I'm
guessing X is anywhere from a dozen to two dozen. Remember that each PC is
well under $2000, and Linux is free, so having a number of PC's is still
quite cheap. I'm assuming that your going to be compiling from C code,
rather than buying precompiled packages, since there isn't much precompiled
software available for Linux yet. Like most Unix systems Linux comes with
a c and c++ compiler and debugger.

If your database is small enough you could stick a copy on each machine
(gig drives are well under $1000 these days).
--
------ Call the skeptic hotline 1-900-666-5555 talk to your own personal .
\    / skeptic 24 hours/day.     Just say no to victimless crimes.      . .
 \  / High quality Linux application development available.            . . .
  \/ Violence is a lousy substitute for sex and drugs.                . . . .

2. Mail Quota

3. Info on OPTICAL DISCS for Unix Vaxen sought

4. Help about sort command

5. Summary (long) on specifying unix server purchase

6. C++ Multithread question

7. Need info on PC/NFS and Unix NFS servers, server/client ratio

8. Booting off Disk & Disks via NFS?

9. seeking advice on laptop purchase

10. Seeking Ethernet Card Purchase Advice

11. Seeking advice on new system purchase

12. Modem Purchase Suggestion Sought

13. Modem Purchasing advise sought ...