Q:
I often use 'prstat ' command to observe current performance of Solaris x86
systems. But I have one question about it. In a report of 'prstat' there is a
column named 'PROCESS/NLWP'. I can't find information to understand purposes of
the column NLWP. Would you explain what this column describes in more detail?
A:
You can find the answer in the man page for it (man prstat).
PROCESS is simply the name of the process.
NLWP is the Number of LWPs used by the process.
Here are two tasks from my system as examples:
You can see that Netscape 7 is multi-threaded and has multiple LWPs.
Although you would never know it from the way it seems to "hang" as one
tab is being updated...
Where the setiathome process uses a single LWP.
LWPs stand for "Light Weight Process" and are part of the implementation
of multi-threading that Solaris uses.
I would suggest looking at the Solaris Internals web site, or buying the
book, to learn a bit more about this:
Q:
Could you please help me with regards to back-ups?
I do not know all the useful commands when doing back ups in UNIX.
A:
The best commands for doing backup and restore are:
ufsdump
ufsrestore
You can use tar or cpio for backing up personal directories and such, but to
backup the system itself you want ufsdump/ufsrestore.
I highly recommend using another disk as your backup device instead of a
tape drive.
Why? Well here is my personal opinion on the subject. Note the word
PERSONAL, please.
Getting up on soapbox...
Because disks are cheap and highly reliable. For safety you can put them in
a removable drawer so they can be taken out of the system and kept off site.
They are also far faster than any tape drive that I have ever used. When you
have a disk crash/failure/corruption happen you simply replace the bad disk
drive with the back, power on the system, and you are ready to go. All that
you would need to do is to try and recover any newer files since the last
backup.
I have never had a tape drive that did not have a high rate of failures. It
runs out of space, the tapes get destroyed (sometimes by the drive itself),
and they are more costly than disk drives and generally hold less data.
I will say that I have not used the latest DLT drives or whatever is out
there these days as the latest and greatest tape solution. But I can backup
200GB of disk onto a cheap IDE disk drive for about $200. I can't imagine
how much the equivalent in tape storage would cost and I will bet it would
not be as fast unless it uses multiple tape drives that are striped in a RAIT
(like RAID, but for tapes) array.
Note that I might use SCSI or Fiber Channel disks for my system drives, but
use IDE disks for backup. The reason is simple: Cost. That would mean you
need to spend more time doing recovery though. If uptime for the system is
critical you should use identical drives for the backup. That way it is a
simple matter of swapping the drive to get the system up and running again.
If uptime is truly critical for you then use a RAID 5 arrangement for your
main storage. For backup you can use another RAID 5 box (more costly) or a
RAID box where you merely stripe the disks (less costly). Again, being able
to swap the backup disks for the failed ones could be worth the extra cost.
Be sure to buy extra disks for any RAID array! Why? Because when they
start to fail you will not be able to find the same type of disks! The
smallest you will be able to find will probably be several times larger.
Plus you want to have them available at the time a drive fails, not several
days later after you order it because a drive failed.
For any kind of backup configuration you should do a cost analysis, of
course. This is insurance, plain and simple. Ask yourself how much it will
cost you if your system is down? Measure it in minute, hours, days, or
whatever is meaningful to you.
My wife's company was down for days trying to use a combination of tape and
internet backups to rebuild a server. It was only a small office of just
over 20 people, but they all sat around (while still drawing salaries) and
did no useful work during those days. One backup disk arrangement like I am
suggesting here would have had them up within a few hours at most with all
their current data to date intact. It cost them something like $15K just to
pay the company that services their systems for their time. It is cheap to
have and easy to use, so consider recovery costs when you plan for backups.
Q:
I built an install server with an image of Solaris 9 on it. I used the last maintenance release
of Solaris 9. Now I want to add the newest patch cluster to this image.
Before I installed the newest patch cluster for the miniroot and it works proper.
I used the patchadd -C command to add the patches to the directory
/install/Solaris_9/Tools/Boot.
After the first setup from this install server the new system wasn't patched!
Is it possible that I must add the patch cluster to the /install/Solaris_9/patches directory?
A:
I was not sure of this myself, so I had to ask the guys that maintain our
install servers here this question. I was not able to find it doing a
quick scan of http://docs.sun.com either.
But the reply I got back says he does believe you are correct. So I did a
bit more searching and found InfoDoc 21571 on SunSolve:
Q:
Will there be an option to purchase the Solaris x86 CD-Images and later get updated CD-Images without paying again (due to increasing H/W support)?
A:
I am a little confused as to how increased hardware support equates to free
CD updates, but here goes...
There is about to be some change in pricing. I can't say exactly what, as
much because I am getting mixed messages as well as because I can't announce
it before it happens. The indications are it will go down from the current
$20 price, which is already pretty darn cheap. I expect to see this change
either later this month or next month, but only time will tell.
Now my understanding is that if you buy a Support contract you would get the
update CDs for the term of that contract. But I am not directly involved in
that so you would have to check to be certain. Also, it might not be Install
media, but Maintainence Update CDs which would only patch your current CDs
and not contain new functionality that was added in subsequent Solaris Update
releases. You would have to talk to a Support representative to see what the
policy is regarding software updates. For Solaris X86 you would want the
Software Only support, most likely.
Look at this page to see what Support Services you can get from Sun:
They list "Software enhancement releases and patches" there for the Software
Support contracts, but I am not certain exactly what that entails. Best to
talk to a live person to be certain.
Now your second question about modems is more complicated, since I don't
really have enough info. Here are the guidelines for modems with Solaris
X86.
The most common type of modem inside PCs these days are WinModems and they
are not supported. They are a lousy modem to use on any multitasking OS,
because they require the CPU to do a lot of the work that really should be
done in hardware. Unless you are running Windows 9x/ME I would avoid them,
as they will drag down the performance of your system.
The next most common would be a PCI modem. The current asy (I am told) is
able to find these. Previous versions did not and you had to figure out
the port and interrupt values and enter them manually into asy.conf to have
Solaris work with them. It should be able to do this on it's own now. But
you probably still have to enter the PCI ID info into /etc/driver_aliases,
since we can't know the ID of every PCI modem out there.
ISA modems should look like a COM1, COM2, etc. device. Note that ISA
devices can NOT share hardware interrupts! So you typically have to disable
the existing COM1 or COM2 when you put the modem in to replace it.
External modems will work fine and I highly recommend them over internal
modems. You then have lights that can tell you something and they will be
able to work on your next computer and the next one and the next... As long
as we have external serial connections. The only caveat is that they do
cost more.
Q:
After installing Solaris 9 x86 (or after applying latest MU to S8/x86) on some
dual CPU systems - mpstat starts showing continuous sys load at about 50% of one
CPU. 'xload' also displays load consistent with mpstat info. What caused this
regression?
A:
You may be running into this bug:
4830628 - MP x86 platforms show load average of 0.3 or greater on idle system
See if the workaround works for you:
psradm -f 0 on a dual cpu machine causes the problem to go away, while not
disabling the cpu.
Q:
How can I access the windows hard drives on my computer when running Solaris 9 x86?
A:
That is very easy. You should read the man page for mount_pcfs, but here
it is in a nutshell:
To mount a Primary FAT partition, which can be FAT12, FAT16, or FAT32 you
can use a mount command like this:
mount -F pcfs /dev/dsk/c0t0d0p0:c /windows-c
To mount a Logical FAT partition, which resides in an extended partition,
you use almost the same command:
mount -F pcfs /dev/dsk/c0t0d0p0:d /windows-d
You may want different directory names, but I find it useful to make them
similar to the Windows drive letters.
I like this method because it is using the same drive letters you would use
in Windows. For a primary partition you can also use the actual partition
number (p1-p4) instead of p0, but why bother? Let Solaris figure it out
for you.
You can also use numbers, where :c becomes :1 and :d becomes :2, but again
why bother? Use the letters you are used to using.
Note that Solaris does not support mounting NTFS partitions at this time.
I am fairly sure Adobe will put up a version of acroread again. Hopefully
the current version and a version of distiller as well, since they have had
a Solaris SPARC version for a long time. Previously only the reader was
available for Solaris X86.
Q:
Can IPSec in Solaris x86 connect to other clients besides another Solaris machine?
My workplace uses Nortel 4500's that authenticate through Radius servers to a Win2k domain.
Netlock has a Sparc client for this purpose but has told me they have no plans for a x86 client.
Can you help?
A:
Funny you should ask that question, because our IPSec interoperability was
something I was just asking about myself for an entirely different reason.
I was told they have done a lot of testing at conferences and such with very
good results. They also test with Windows regularly and I am told they plan
on writing up something about that, but that it works with Solaris just fine.
Now I am not sure just what Netlock gives you, but it is possible that you
don't need it with Solaris 9 now that the OS supports IPSec. If you have
Solaris 9 now you should test it and see if you still need Netlock.
If it does add some functionality beyond bare IPSec that you need I would
ask them why they are not doing an X86 version. It really should be just a
fairly simply recompile, unless byte ordering matters to their code. We are
selling a fair number of X86 systems and that quantity will only get larger
as time goes on. They are limitting their market if they ignore Solaris X86.
The current version appears to be 3.22 on that site.
As for MIME, I am guessing you are talking about MIME types? Solaris
does support them and has for quite a while. You have .mime.types file in
your home directory. Both dtmail (CDE) and Netscape support MIME types in
E-mail attachments.
Q:
We are trying to install Solaris9 (x86) on a Dell 2650.
Our problem is that we have no drivers for the LSI Logic (formerly AMI) PERC3/DC controller.
Thus the drives are not recognized.
Dell has been contacted and they do not provide the driver (only firmware).
I could not find a driver via LSI either.
I noticed your comment about the Adaptec SCSI driver not being available with Solaris9 4/03.
Can you help? This problem will influence our OS direction!
A:
Well I can say that we are trying to work with LSI to get the mega driver
back into production, as Dell is an important partner of ours and this is
a popular card for Solaris users.
However, they seem to have withdrawn their Solaris driver sometime after we
had put Solaris X86 on hold and have not yet brought it back. There was a
lot of discussion about problems with the mega driver they had posted there
in the Yahoo group SolarisOnIntel:
A Google search will turn up several references. Based on what I read there
the 2.18 version may not be the ideal one to use. Some were using the 2.17
and others the 2.19 versions. Unfortunately, all I have myself is the 2.18
version.
You will note that the SCO Unix 5.x is also 2.18 there. This makes some
sense, since they use a similar driver model. That is why we recently made
a purchase of their driver source code, which you may have heard in the news.
Unfortunately, SCO's Unix 5.x is not identical to Solaris in how you would
apply the driver and it probably is also not identical at the binary API
level either. So I would not recommend trying to use that version.
I would suggest joining the Yahoo group and asking if someone there has a
version that works for them that you can use until LSI is back on board with
Solaris X86.
We are working to get other RAID solutions to Solaris X86 ASAP. There are
several Fiber Channel solutions out there as well. Note that a Solaris 8
driver will usually work just fine on Solaris 9. I have done that several
times myself. You might try searching for Solaris RAID drivers in Google
or your search engine of choice.
In my lab I am using a Sun 3310 SCSI external RAID box. It works with any
SCSI card and can be configured by serial port or the network interface. It
is yet another option to consider.
Q:
I need any and all info on Solaris 9 x86 Real Time OS performance;
e.g. jitter, context switch times, etc. (the Full Monty:)
I need to know specifically which x86 boards run the fastest.
My inquiry is in regards to a Real Time deal with Raytheon.
They are currently going with IBM Power at 4 flts/clk vs our Sparc 2 flts/clk.
Our Sparc FFT performance stinks to put it mildly. Therefore we need x86 numbers.
Also if you can get me "time" on ANY x86 Benchmark System I would sell you my baby brother:)
I need to run the public domain FFTW package from MIT.
A:
Sorry, but I don't have any specific numbers handy for you. I would say you
can look here for some:
That might at least help you determine which systems run faster. I do know
that our new V60x and V65x are about as fast an X86 system as you can find
in a dual processor system. We are working on qualifying some quad CPU
systems as well.
I do know that for raw integer and floating point you will find that the X86
CPUs are much faster than the current SPARC systems. Many people that do
some serious number crunching with Solaris have told me that, which is why
they have been asking for us to bring back our Fortran, the language of
choice for many serious number crunchers.
The reason for X86 speed over SPARC is simple:
1) Faster clock rate
2) Competition between manufacturers for performance
Now for raw sustained throughput on multiple CPUs I will choose a SPARC every
single time. The X86 CPUs do not work as well in a multiprocessor situation.
But for 1-2 CPUs, and maybe for up to 4 CPUs, you will probably get better
numeric speed from an X86 system at this time. Our new faster multi-threaded
CPUs that are coming out will certainly close that gap, so this will have to
be constantly re-evaluated.
I have BCC'd someone here at Sun that does benchmarks for both platforms so
he may be able to help you get more specific information that will help you.
As to your question about the realmode performance I simply do not know. I
can quote from the Solaris Internals course that the "dispatch latency for
the highest priority process is kept between 2 and milliseconds depending
on the architecture". You might also take a look at the Real-Time Dispatch
Parameter Table in that same document.
There is also a white paper in the course materials:
Realtime Scheduling in SunOS 5.0 by Sandeep Khanna, Michael Sebree, and
John Zolnowsky
I found many references to it when I did a search for that title. Since it
is old you might want to look at newer papers that refer to it, as well as
the original one.
Q:
After building a Bootable Solaris CD-Rom, is there any way to get the DCA to _not_ pop up and bother me?
I know the platform I'm installing on, and the defaults work great with no ITU's, I just want it to do it's job and go away.
How do I do this?
A:
This is a very interesting question and I can give you a partial answer to
it right now. We are currently working on a better solution for hand-off
installations over the network that would also apply if you build custom CDs
for your installation. This was a topic of discussion on an internal alias
and so I saved all the juicy info I could about it, because I don't think it
is documented as well as it should be yet.
Get a cup of coffee or whatever, as this will be a long reply...
This method works when you are installing multiple systems that are exactly
identical to each other. All you need is one to create the info file for
you to use.
To avoid being prompted by kdmconfig to configure your video you can put
these lines in your sysidcfg file:
Now for something that was not documented at all. How to write a bootconf
script file. Here is a sample script that worked for me on a test system:
F2
F2
SELECT 1
F2
You can enable this by editting bootenv.rc in the /solaris directory on a
boot floppy and changing the setting for 'confflags'. For a boot floppy
this will already be set to "-f". You can change it to either:
"-f -Ptest.scr"
or "-f -ptest.scr"
Either will cause the file test.scr to be read from the root directory of the
floppy and used in place of keyboard input. The difference between -P and -p
is that the uppercase option allows you to see the screens as you normally
would. The lowercase one disables screen output after the initial screen
where it says "Running Configuration Assistant". You can put the file in a
subdirectory if you wish, but the default with no path supplied is to load
from the root directory of the boot device.
So to make an automatic script for your system I would follow these steps:
Go through the screens manually and write down every keystroke you use
until you get to the final screen that lists your possible boot devices.
Count down the list, starting with 0, to determine which argument to use
for your SELECT command.
Write your script and test it using the -P switch so you can watch for
any problems as it executes. Personally, I would leave it at this, just
so the display of each screen can be used to show what phase it is in.
If you wish to hide those screens change it to the -p switch after you
have determined that your script works correctly.
To skip past ACPI warning messages you might have to add some ENTER lines
as well. But for a simple system where you just have to hit and to
select your boot device this is all that you need. About the only thing
that would normally need to change would be the SELECT line above, so you
select the particular device for your system that you are using to install
from.
When installing multiple identical systems in this manner you must be sure
that everything is identical in them, down to the very slot that any card
is plugged into that might appear as a legitimate boot device and any disks
that are attached to the system. If you do not then the SELECT line may not
work, because the order of the install choices can change.
Please note that this is an undocumented and unsupported feature of the DCA
environment, so there is no guarantee that it will work in older or future
versions. I personally tested it with Solaris 9 4/03 myself, so it does work
there. It has many other features, but this is all you will need for the
simple purpose of automating the DCA.
Q:
Is there a way to implement password aging under NIS?
A:
This is straying away from my area of expertise, but I did some searching
on http://docs.sun.com and was led to the pages about PAM. Try looking at
this page:
While NIS does not support aging, it seems that PAM can add that capability.
I really can't say, as I have only just read it myself and have not tried it.
It might be worth looking into LDAP later, as I think that is going to end
up replacing both NIS and NIS+ eventually.
Q:
When I run /usr/X/bin/xdpyinfo it shows backing-store and save-uppers as NO.
How can I change this to YES?
A:
Well our X86 display adapters are handled a bit differently than those for
SPARC, so I had to check on my own system.
On my system I have an ATI Rage Pro XL and it does have these set to YES.
I can only guess that your particular driver does not support those type
of settings being set to YES. I would be curious to know the make of your
system and the video card it has in it. I would also like to see a copy of
your /etc/openwin/server/etc/OWconfig file. That will tell me how it is
configured as far as drivers your are using, etc.
Q:
This question is regarding a problem using Live Upgrade 2.0 (LU) on x86 system.
I've got a single 4.2 GB IDE drive. During installation I partitioned the Solaris slice
in the following way:
/ - 1.07 GB
swap - 256 MB
/var - 512 MB
/export/home - 200 MB
I also reserved 1.3 GB for ABE to
hold root partition. The Solaris slice is occupying 55% of the total size of the
hard drive and the rest of HD I left unpartitioned planning to install second OS
for multyboot configuration.
Now when I am trying to use LU, I have enough space to hold root partition
( using 1.3 GB left unassigned on the Solaris slice), but I cannot figure out how to allocate unpartitioned
(raw) disk outside of Solaris slice for alternate /var partition.
Using fdisk I created another Solaris slice
which takes another 20 % of total space on HD, but I cannot format it for some
reasons and LU doesn't see that second slice I created (and because I have
enough space only for / slice and I am short about 600 MB for /var slice I am
stuck).
The question is, if I want to use LU on the x86 system, how can I make
unpartitioned (raw) hard drive available to Live Upgrade?
I bet Solaris 2.6, for
example, had not been partitioned for Live Upgrade in mind during installation.
So the obvious solution is to add second HD big enough to hold all existing
slices, and there is should be a way to make that second HD visible to LU. I
tried to do that kind of operation (adding second SCSI HD and using it for ABE)
on the SPARC system (sun4u) and I had no problems. It looks like only x86 system
I am having problem with.
A:
Well I really have not used Live Upgrade myself, but I can speak about the
problems you had with the second Solaris partition.
Solaris only supports a single Solaris fdisk partition per disk. I think
the way it searches for the one it will use is first it looks for an active
Solaris partition. If none are active then it uses the first table entry
with a Solaris fdisk partition ID (which is also used by Linux for swapping).
So I would try adding the second hard disk and making a Solaris partition
on it. Not knowing how Live Upgrade wants to use a disk you might try both
with and without a Solaris partition and maybe with a Solaris file system
slice in the Solaris partition.
Unfortunately we do not have a tool that grows (or shrinks) a Solaris fdisk
partition. Nor do I know of any third party tool that can do so.
