How to configure Sun15K console remotely

[Krishna]

I have several question on Solaris9 Sun 15K, since I am new to it.

1. How to configure cosole login remotely on Sun 15K?

2. How to configure domains and what is the maximum amount of domain
permitted
on Sun 15K

3. Automount Process though NFS mount

4. IF the 15K is slow what is the usual procedure to follow to
identify what is the problem.

5.Is there any change on SMS from the 10K?

Thanks in advance

rkrish

[Elias]

Krishna wrote:

> I have several question on Solaris9 Sun 15K, since I am new to it.
>
> 1. How to configure cosole login remotely on Sun 15K?
>
> 2. How to configure domains and what is the maximum amount of domain
> permitted
> on Sun 15K
>
> 3. Automount Process though NFS mount
>
> 4. IF the 15K is slow what is the usual procedure to follow to
> identify what is the problem.
>
> 5.Is there any change on SMS from the 10K?
>
> Thanks in advance
>
> rkrish

1. Remote console login is done through the system controller either
by ethernet or serial console through a terminal server.

2. Domains are configured via the platform shell on the system
controller. You can have up to 18 domains or the number of installed
system boards if you don't have a fully populated system.

3. /etc/vfstab see
[url]http://docs.sun.com/db/doc/806-4073/6jd67r9in?a=view#fsmount-29904[/url]

4. Figure out what the bottleneck is for your configuration and
application workload and fix the bottleneck.

5. Yes.

I realize that these answers are vague but you didn't provide much detail.

More information can be found at:
[url]http://docs.sun.com/db/prod/servers.15k#hic[/url]


Elias

[Scott Howard]

Krishna <rkrishp@hotmail.com> wrote:

> I have several question on Solaris9 Sun 15K, since I am new to it.

I think what you really need to consider is
[url]http://training.sun.com/US/catalog/courses/ES-421.html[/url]

Seriously, the SF15K is a big, complex machine, which is generally going
to be used for mission-critical situations - leting sysadmins who are not
15K trained admin them is not a good way to run things.

No matter how well you know Sun gear/Solaris (even if you know E10K)
the 15K is significantly different - definitely different enough that
you should condsider some form of formal training before admin'ing one.

Scott

[Remco]

I fully agree with Scott!

As a former instructor ... and I see those Questions...
Attend the ES 421 from Sun and mainly all your questions and more will be
answered.

And until you have followed that course... stay off the SF15k.





"Scott Howard" <scott@hunterlink.net.au> wrote in message
news:1059360882.555531@docbert...

> Krishna <rkrishp@hotmail.com> wrote:

> > I have several question on Solaris9 Sun 15K, since I am new to it.

>
> I think what you really need to consider is
> [url]http://training.sun.com/US/catalog/courses/ES-421.html[/url]
>
> Seriously, the SF15K is a big, complex machine, which is generally going
> to be used for mission-critical situations - leting sysadmins who are not
> 15K trained admin them is not a good way to run things.
>
> No matter how well you know Sun gear/Solaris (even if you know E10K)
> the 15K is significantly different - definitely different enough that
> you should condsider some form of formal training before admin'ing one.
>
> Scott

[Sam N]

"Remco" <Boarddude@zonnet.nl> wrote in message
news:3f24bd82$0$45372$1b62eedf@news.wanadoo.nl...

> I fully agree with Scott!
>
> As a former instructor ... and I see those Questions...
> Attend the ES 421 from Sun and mainly all your questions and more will be
> answered.
>
> And until you have followed that course... stay off the SF15k.
>
>
>
>
>
> "Scott Howard" <scott@hunterlink.net.au> wrote in message
> news:1059360882.555531@docbert...

> > Krishna <rkrishp@hotmail.com> wrote:

> > > I have several question on Solaris9 Sun 15K, since I am new to it.

> >
> > I think what you really need to consider is
> > [url]http://training.sun.com/US/catalog/courses/ES-421.html[/url]
> >
> > Seriously, the SF15K is a big, complex machine, which is generally going
> > to be used for mission-critical situations - leting sysadmins who are

not

> > 15K trained admin them is not a good way to run things.
> >
> > No matter how well you know Sun gear/Solaris (even if you know E10K)
> > the 15K is significantly different - definitely different enough that
> > you should condsider some form of formal training before admin'ing one.
> >
> > Scott

>
>

Couldn't agree more - ES-421 is what you need. Trust me, the investment more
than pays for itself.

cheers

Sam N

[Philip Brown]

[prior post has expired, so have to reply to reply...]


On Mon, 28 Jul 2003 08:10:59 +0200, [email]Boarddude@zonnet.nl[/email] wrote:

>I fully agree with Scott!
>...
>"Scott Howard" <scott@hunterlink.net.au> wrote in message
>news:1059360882.555531@docbert...

>>...
>> Seriously, the SF15K is a big, complex machine, which is generally going
>> to be used for mission-critical situations - leting sysadmins who are not
>> 15K trained admin them is not a good way to run things.

Except that the 15k is NOT really mission-critical. or not as much as it
SHOULD be. This really pisses me off.

Even in a multi-million dollar box, sun still crashes the ENTIRE BOX, if
one stinking CPU board goes down.
This is NOT mission-critical design. IBM have been doing it right for
years. What should happen is that as long as it isnt a kernel board,
just the processes running on that particular board should SIGBUS,
some nasty complaints should be logged in /var/adm/messages,
and the board should be redlisted, but the system should KEEP RUNNING, dammit!!

Everyone who has a 15k system, should take the time RIGHT NOW, to file a
priority 1 bug against this issue, and escalate it until you have an actual
architecture/kernel engineer acknowlege the issue in the ticket.
For X million dollars, times 2000 customers, sun should be treating the
mission-critical business a whole lot better.



--
[url]http://www.blastwave.org/[/url] for solaris pre-packaged binaries with pkg-get
Organized by the author of pkg-get
[Trim the no-bots from my address to reply to me by email!]
S.1618 [url]http://thomas.loc.gov/cgi-bin/bdquery/z?d105:SN01618:@@@D[/url]
[url]http://www.spamlaws.com/state/ca1.html[/url]

[Philip Brown]

Oops, forgot this most important bit...


On 31 Jul 2003 14:08:00 GMT, [email]scott@hunterlink.net.au[/email] wrote:

>...
>If the
>error isn't detected before it becomes fatal (and often it's not
>possible to) then Solaris takes whatever action is required to maintain
>data integrity - either killing processes and rebooting (if the problem
>occured in userland),...

Forced rebooting needlessly kills uptime/service availability.
Particularly on a fully loaded E15k.
Sane people have these things clustered.
(or just really well-wrapped with scripts).

In a clustered situation, the cluster software has the opportunity to try
simply restarting the application if it crashes.

So if oracle goes down, due to a failed board+process_kill, it could try just
restarting oracle. Potentially a 10-minute downtime, instead of a 20 minute
downtime.

If just a few client connections go down, because of process_kill, then
oracle STAYS UP, with just a few clients kicked off.
**0** minute downtime, instead of a 20-minute downtime.

But no, the E15k domain has to REBOOT. Costing companies
MILLIONS OF DOLLARS, in the case of milion-dollar-per-minute environments,
**NEEDLESSLY**, because sun doesnt think it's important enough to give the
customer a choice on whether their multi-million-dollar hardware reboots
on system board errors.



--
[url]http://www.blastwave.org/[/url] for solaris pre-packaged binaries with pkg-get
Organized by the author of pkg-get
[Trim the no-bots from my address to reply to me by email!]
S.1618 [url]http://thomas.loc.gov/cgi-bin/bdquery/z?d105:SN01618:@@@D[/url]
[url]http://www.spamlaws.com/state/ca1.html[/url]

[Anthony Mandic]

Philip Brown wrote:

> So if oracle goes down, due to a failed board+process_kill,

Oracle doesn't need a hardware failure to go down.

> But no, the E15k domain has to REBOOT. Costing companies
> MILLIONS OF DOLLARS, in the case of milion-dollar-per-minute environments,
> **NEEDLESSLY**, because sun doesnt think it's important enough to give the
> customer a choice on whether their multi-million-dollar hardware reboots
> on system board errors.

Million-dollar-per-minute environment? Is that possible yet?

-am © 2003

[Philip Brown]

On 1 Aug 2003 09:10:02 GMT, [email]scott@hunterlink.net.au[/email] wrote:

>Philip Brown <phil+s3@bolthole.no-bots.com> wrote:

>> Nice theory. But again, theory != practice.
>> In practice, we have found this functionality you have described above, to
>> be nonexistant. We have had multiple E15k domain crashes in the last few
>> months, due to failed system boards.

>
>What version of SMS are you running? What SMS patches? What KJP on
>the domain?

I notice you didnt address my issue of "needlessly rebooting causes
at least 10 minutes of unneccessary downtime", nor the issue of giving the
customer the choice of action in those situations.

Whether or not we are fully patched, does not fix those more important
issues.



--
[url]http://www.blastwave.org/[/url] for solaris pre-packaged binaries with pkg-get
Organized by the author of pkg-get
[Trim the no-bots from my address to reply to me by email!]
S.1618 [url]http://thomas.loc.gov/cgi-bin/bdquery/z?d105:SN01618:@@@D[/url]
[url]http://www.spamlaws.com/state/ca1.html[/url]

[Philip Brown]

On Fri, 01 Aug 2003 16:43:45 +1000, [email]q0@hotmail.com[/email] wrote:

>Philip Brown wrote:

>> But no, the E15k domain has to REBOOT. Costing companies
>> MILLIONS OF DOLLARS, in the case of milion-dollar-per-minute environments,
>> **NEEDLESSLY**, because sun doesnt think it's important enough to give the
>> customer a choice on whether their multi-million-dollar hardware reboots
>> on system board errors.

>
> Million-dollar-per-minute environment? Is that possible yet?

Thats what our (internal business) customer tells us. You know, the people
we actually administer the machine FOR.


--
[url]http://www.blastwave.org/[/url] for solaris pre-packaged binaries with pkg-get
Organized by the author of pkg-get
[Trim the no-bots from my address to reply to me by email!]
S.1618 [url]http://thomas.loc.gov/cgi-bin/bdquery/z?d105:SN01618:@@@D[/url]
[url]http://www.spamlaws.com/state/ca1.html[/url]

[Anthony Mandic]

Philip Brown wrote:

> > Million-dollar-per-minute environment? Is that possible yet?

>
> Thats what our (internal business) customer tells us. You know, the people
> we actually administer the machine FOR.

They might be exaggerating but, looking at the figures, its $60M
per hour. With, say $100 p.h. on average per employee, that's
about 600,000 employees connected to it. At $500 p.h. its 120,000
employees. With only a handful employees, its mucho dinero. 10 minutes
and I'm outa there. So, how busy are the systems usually?

-am © 2003

[Lon Stowell]

Philip Brown wrote:

> On Sat, 02 Aug 2003 15:41:03 +1000, [email]qh@hotmail.com[/email] wrote:

>>Philip Brown wrote:
>>

>>> > Million-dollar-per-minute environment? Is that possible yet?
>>>
>>> Thats what our (internal business) customer tells us. You know, the people
>>> we actually administer the machine FOR.

>>
>> They might be exaggerating but, looking at the figures, its $60M
>> per hour. With, say $100 p.h. on average per employee, that's
>> about 600,000 employees connected to it.

>
> You're using the wrong metrics.
> You're thinking, "if employees cant connect, we are wasting money by paying
> employees to do nothing for an hour".
>
> The actual cost is lost of **customer business** for that hour

An even more accurate metric isn't just the customer business you
lose for that hour. Add in the continuing loss of customer business
from those customers who lose confidence in your operation because
your systems are down. e.g. a webserver down for 1 minute at a
company that claimed to sell highly reliable web services would
mean I'd probably look elsewhere.

[Anthony Mandic]

Philip Brown wrote:

> You're using the wrong metrics.

Yes, I probably am, but it would depend on the nature of the business.

> The actual cost is lost of **customer business** for that hour

That would only apply to customer facing businesses. Manufacturing
would be different. But businesses trading with customers would
certainly be affected. However, how long is the actual average
down time? If its only a few minutes, it easy enough to keep the
customer talking on the phone, or take their order down on paper
and process it later. Another technique is to switch the phone
systems over to busy, and let the callers queue or ring back.
This brings to mind some old experiences. Some problems aren't
computer related but still affect business income. At an old site,
the PABXs went down during a power loss. The computers were on
UPS and so weren't affected but the telephone operators couldn't
take any calls. Same problem when the phone lines were cut out
in the street.

-am © 2003

[Anthony Mandic]

Lon Stowell wrote:

> An even more accurate metric isn't just the customer business you
> lose for that hour. Add in the continuing loss of customer business
> from those customers who lose confidence in your operation because
> your systems are down. e.g. a webserver down for 1 minute at a
> company that claimed to sell highly reliable web services would
> mean I'd probably look elsewhere.

Depends on the nature of the business. If you're a monopoly, the
customers have no choice but to deal with you. If you're not,
your reputation would carry you to a certain extent. If your
business was best of breed, they'd still be more likely to keep
doing business with you (if only because dealing with the
competition isn't desirable).

Something like a webserver is highly visible and obvious but,
when a customer can't really "see" what's going on, there's no
real reason to tell them specifics. With the example I cited in
my other post in this thread, the phone operators would switch
to pen and paper if the computers went down. The customers never
knew unless they wanted to make enquiries (in which case the
procedure was to switch them to another operator who would
appear to be busy and put customers in a queue). Phone systems
failing would give a customer the impression that the problem
was with the phone service rather than the business, so that
was rarely a problem.

-am © 2003

[Philip Brown]

On Sat, 02 Aug 2003 18:50:22 GMT, [email]lon.stowell@comcast.net[/email] wrote:

>Philip Brown wrote:

>> You're using the wrong metrics.
>> You're thinking, "if employees cant connect, we are wasting money by paying
>> employees to do nothing for an hour".
>>
>> The actual cost is lost of **customer business** for that hour

>
> An even more accurate metric isn't just the customer business you
> lose for that hour. Add in the continuing loss of customer business
> from those customers who lose confidence in your operation because
> your systems are down. e.g. a webserver down for 1 minute at a
> company that claimed to sell highly reliable web services would
> mean I'd probably look elsewhere.

Or for that matter, a company that claimed to sell highly reliable servers.



--
[url]http://www.blastwave.org/[/url] for solaris pre-packaged binaries with pkg-get
Organized by the author of pkg-get
[Trim the no-bots from my address to reply to me by email!]
S.1618 [url]http://thomas.loc.gov/cgi-bin/bdquery/z?d105:SN01618:@@@D[/url]
[url]http://www.spamlaws.com/state/ca1.html[/url]

[Anthony Mandic]

Philip Brown wrote:

> Or for that matter, a company that claimed to sell highly reliable servers.

Dare we ask why, with the amounts involved, there was no HA or
replication solution in place?

-am © 2003

[Anthony Mandic]

Philip Brown wrote:

> HA STILL GIVES YOU DOWNTIME. You're missing the point.

OK, OK.

> and oracle 9iRaq hasnt been certified for the application involved,
> before you ask. and no, I cant tell you what the application is.

I still observe the DBAs bitching about RAQ, so it will probably
be some time ...

-am © 2003