Letters to the editor

1. SunWorld Online is not owned by Sun Microsystems. It is an independent publication about Sun-related computing issues.

2. I write for many publications, so I am not even an employee of SunWorld, much less Sun.

3. The deal between Sun and Apple is not yet done. Apple has not agreed to the buy-out offer, yet.

4. Whatever happens if Sun buys Apple, it could be worse. Oracle or IBM could buy Apple.

My guess is that Apple computers and the MacOS will live for many more years. However, what would you think of running the MacOS and applications on an UltraSPARC box?

How about some CAD coverage?

Editors:
As a long-time reader of your zine (both paper and electronic), I have always wanted to see an in-depth article on Suns and CAD. For example a comparison of Sun's graphics technology (ZX, TZX, Creator) and maybe some discussion of performance tuning for CAD work (A mix of floating and integer performance). Also, how about a look at the newest hyperSPARC processors (133 and 150) and how they perform in SPARC 10 and 20 configurations. Just a thought...

--Jeff Zemsky, CAD Administrator, Garden Way Inc.

PS - Love the new e-zine format. It's timely and informative (and unlike some of the other Unix magazines -- the articles deal with today's Unix news, not products and news from a year ago :)

Java questions and comments

Editors:
I would like to study HTML or anything about developing Web pages. I would like to get a copy of Java. Thank you.
--(name and firm indeterminate)

Editors:
In one of you past issue you had a list of Java applets people wanted to see. Look at http://smc.vnet.net/javalamp.html for a Java Lava Lamp generated with Mathematica and a Java animation applet. Details on how this was done are given.

--Steve Christensen, (firm indeterminate)

Editors:
First off, let me say that finding this GEM of a place is very refreshing and very informative.

I've been preaching Java and HotJava since first downloading it in late August. Since then, I was amazed by the possibilities but since having no background in programming, I was at a loss. Recently, I've managed to learn HTML and VRML thus was I entrigued by Java that I returned to embedding distributed applets into my homepage. I currently have two working and am experimenting with modifying them by exchanging the GIF involved and sound file to create a "new" applet. Perhaps there is a story behind this? I shall let you know how I've made out.

--(name and firm indeterminate)

Dear people from Sun:
I think that Java, together with the Net, is one of the most important inventions. But you people are in the duty to make it easier to get technical information about the Java language [out to people]. I, for myself, want to write a Java interpreter, which shouldn't be a big deal. But I have a hard time to find the required information.

Another thing is the global view. I am from Europe, more precisely, from Germany. Most people there still think that the Java concept is more "cold coffee" than hot Java. I think it is best if we can convince people and companies all over the world to support Java. Otherwise we maybe end up with different standards again. And I think that's not in the interest of the user. (Like we have different standards for VCR and TV.) I would be more than glad to be able to support Java and your concept, because I think thats the future, and that's the solution, and you have a very realistic opinion.

But I need also a little support from you. Or I would be more than happy and pleased to work on projects together with you. My background is in industrial real-time controlling, and data remote controlling, but I am thrilled by the possibilities I see.

--Mar Ricke, (firm indeterminate)

Needs help from Sun

Editors:
I am a system administrator of some Sun workstations in Hong Kong. I have a lot of questions on system administration. May I raise those questions to you and get some comment on those issues? If I cannot leave the questions here, where can I find the support from Sun? Please give me some advices!!!

--Lawrence Ngai, (firm indeterminate)

Thoughts on Polish Internet situation

Nelson ISP benchmarks on right track

Editors:
As competition heats up for the internet access dollar, I have heard impossible service claims being made. I feel Neal Nelson is on the right track developing a benchmarking tool for ISP evaluation. As a professional information systems auditor, criteria for evaluating the quality of ISP's will help consumers (both commercial and retail) decide which provider will best satisfy their needs. Currently, this decision is a hit/miss proposition. I don't support governmental regulation of the service, however self-regulation doesn't seem likely. ISP's need qualified independent third party assurance using a common set of standards. This may seem visionary, but will help clear the muddy waters for the demanding consumer.
--Randall W. Riley, CISA, Orlando, FL

Career Advisor: Java detail flub

Editors:
In the Resources section of the February 1996 Career Advisor, it incorrectly states: "To see Java applets unfold on your screen you must have Windows 95, NT, or Solaris. These are the only operating systems which support Netscape 2.0, the only browser which currently runs Java beta code." In fact, there are several other versions of Netscape 2.0 that are Java enabled. At the very least, the Linux version runs Java applets (I am currently using it).

I think the writer is confusing Netscape 2.0 with the Java JDK, which is currently officially available from Sun only on the platforms mentioned. However, quite a few other ports of the JDK are available unofficially for other platforms, so to say that one must have Windows 95, NT, or Solaris to view Java applets is erroneous in any respect.

--(name and firm indeterminate)

Hello, Edgar:
With Your article in SunWorld Online you helped me very much in being confident in staying with Java in the future. Do you know U.S. companies in demand of Java (or other) software (or hardware) documentation translation to the German language? I could get and send the translation data over the Internet. I could even stay at the company in the U.S. for a time.

--Albert Kastl, (firm indeterminate)

Response to an earlier Career Advisor question

A further excellent resource for Systems Administration issues is SAGE, the Systems Administrators Guild. SAGE publishes an annual Salary Survey which is broken down by region, and the SAGE jobs mailing list also has a fair amount of traffic.

For more information on Pencom Systems Administration Services, including an interactive salary guide and job listings, please see http://www.pencomsi.com/careerhome.html.

For more information on SAGE, please see http://www.sage.usenix.org/sage. Mr. Saadi is a far better writer than I, and his previous columns in the back issues of SunWorld Online contain many references and much information which should be helpful.
--Stephen P Potter, Member, Pencom Systems Administration Services

A Java Career

Hal's SysAdmin FAQ

Could you also please tell me how can I take file system (root, usr, etc.) backups on tapes safely in multiuser mode which can then be used for restoring a system after its crash? How can I safely and quickly restore a sun system after its crash?

Lot of thanks in advance.
--Babu Koshy, Tokyo

The second part of the backup problem is if you are experiencing disk failures, or worry about quick recovery, the best solution is to not have to backup from tape at all. Use a disk mirroring or volume management product (like Solstice DiskSuite) to keep a "live" copy of the data online, and you can switch away from a failed disk with the system running. Disks are inexpensive enough now that this may prove more cost effective than building a deep nest of incremental backup scripts.

To learn more about system administration, let me suggest the O'Reilly system administration book series. They are available directly from O'Reilly, or translated into several languages through local distributors. Evi Nemeth's System Administration Handbook is also quite good, especially the second edition that covers Solaris as compared to other operating systems. For up-to-date information, be sure to join the sun-managers mailing list (send e-mail to sun-managers-request@ra.mcs.anl.gov, with a body of add). A frequently asked questions (FAQ) list for sun-managers is available from ftp://ra.mcs.anl.gov/sun-managers/faq. There is also an archive of all summary responses posted to the alias available for ftp from aurora.latech.edu

RAID 5 and storage questions

Dear Brian Wong:
I am looking for a disk subsystem for Sun Microsystems computers that I can connect two Sun SPARCcenter 1000s to. One of them is Primary Server running Sybase SQLServer 10.x; the other is a Online Hot Standby, In case of the Primary Server failure, it just needs to reconnect to the other host to replace its function.

--Otto Howe System, Administrator HK Star Internet, Kowloon, Hong Kong

Dear Brian:
I have searched high and low for what would seem like a simple answer, but I can't seem to find it. Then I found your article on RAID technology and thus your name. I hope you can help.

I understand RAID 5. The problem has risen when that understanding is mixed with the use of Volume Manager. The following explains:

I used Sun's Volume Manager (Version 2.1) to select and create a RAID 5 volume for 7 one-gig drives in our SPARC Disk Array Model 100. I let Volume Manager do its thing and the result was a RAID 5 that was about half the size of the total space available. For example: Out of 7 gigs I had about 3.0 being used by the RAID 5 and being reported by df. The Volume Manager reported that there was XXXXXXXXX number of blocks available. So I said gee, I would like to use all that space in the RAID. So I "grew" the volume and the disk until there were no more blocks available being reported.

Concern: Did I mess up Volume Manager's RAID 5? Did it leave that much space unused for a reason? I would think not. Or is Volume Manager designed to create only a fraction of the blocks available into a RAID 5 and then you do the rest? Each time I told Volume Manager to grow the disk and the volume, it would only do a fraction at a time. Example: with 5,000,000 blocks available it would grow 1,000,000, then I would grow 4,000,000 and get another 1,000,000, etc., etc., until I finally had no more blocks. (The above numbers are not accurate, they are only used for example.)

Now I'm worried I messed things up. Any facts, ideas, suggestions, insights on this would be greatly appreciated.
--George Sullivan, (firm indeterminate)

Dear Brian:

In your article on RAID you stated that in RAID 3 the parity disk becomes the bottleneck. However, if all the drives in the RAID rank are synchronized so that both data and parity were read or written at the same time, the parity disk couldn't become a bottleneck.

Also, if a stripe per disk larger than a byte was used, say 1024 bytes, and four data disks were in the rank, wouldn't this result in acceptable random performance? For every I/O, 4096 bytes would be transmitted to the host, which would meet the requirements of most online systems.
--Rick Sewell, (firm indeterminate)

This is why essentially all "RAID 3" implementations are actually RAID 5 under the covers.

I disagree with your contention regarding random performance, for the reasons stated above. RAID 3 fills a very narrow niche, and one that can be filled adequately by RAID 5.

Dear Brian:

We use a program call pvfs that was written by Pinnacle Micro to handle their jukeboxes. The program was working perfectly under Solaris 2.4. But under 2.5, we met the following error: "NFS server not responding, broken pipe" Do you know any quick fix to the problem? Pinnacle Micro is absolutely helpless. Thanks!
--Hai-Ru Chang, (firm indeterminate)

Emacs vs. vi will never die

Your article was an eye opener. I always thought that vi was all we had on the Unix. But isn't it a vicious circle that because we have comparatively few users of emacs vis-a-vis vi, we also have very few people/resources to turn to when we are learning emacs if it is installed on our machine? Maybe you could come out with a tutorial, "Learning Emacs." That would be great for would-be vi quitters.
--Jaihar Murlis, (firm indeterminate)

Dear Chuck Musciano:
I use Joe, a simple editor based on PC editors using the Wordstar control sequences. (Yes I know, Emacs can be configured to do that, I've thought about using it just to confuse the Emacs people I work with.)

Another choice is Crisp, another PC-like editor which mimics Brief. PCs sell so much because they're easier to use; there might be something to that! These are some pretty good editors for Unix. Joe is simple, has macro abilities, and can be reconfigured; most of which I ignore.

Emacs is too large with too many features. I would some day determine that I needed to do just "this" to the document I was editing and instead of doing it by hand in twenty minutes, I would spend three hours figuring out how to make Emacs do it for me, write into some little macro I would name something very unique, and when I would have to do that task again, spend 25 minutes finding the name of the macro instead of doing the task.

And why do I want to browse the Web or receive mail in my editor??

Although I use vi, I really enjoyed your article -- by the way do you know if there is a drop-in version of vi for VMS? I am stuck these days using VAX, and need this, but do not want to invest in learning VAX/VMS to be able to compile it if there is source code ready for VMS but not the executable.
--Todd Sabuncu, (firm indeterminate)

Letters, our performance columnist gets Letters

Dear Adrian:

I'm running a very large web server and I seem to have the directory cache "go to amber" (under ruletool) for poor DNLC hitrates. I increased ncsize to 34000 on a : *) 512 MB 4-CPU SPARC 1000 running 2.4

Is that the highest ncsize can go? Would tuning somewhere else help?
--(name and firm indeterminate)

Dear Adrian:

There is some confusion about which target sd0 in the iostat output refers to. Some people here say it is target 3 (Internal drive) while others say it is target 0. Which would be correct? The same quandary exists with the vmstat disk fields, s0 and s3. Which target does each field refer to? Thanks and I am going to buy your book this afternoon!

--Jeanne Brennan, brennan@hou.moc.com

% /opt/RICHPse/examples/disks.se sd0 -> c0t0d0 sd2 -> c0t2d0 sd3 -> c0t3d0

Dear Adrian:
Are the SymbEL Interpreter, extensions, and Rules/Tools supported on Solaris 2.5 Driver Update 1 for X86? If so:

1. Can you use the packages from the September 1995 articles?
2. Do you need to download new code? Where is it?

Dear Adrian:

We are busy porting a real-time application from Linux to Solaris x86, and are experiencing problems with regard to memory.

We load about half of our physical memory with data (30 MB), and even though there should be plenty memory available, we experience lots of paging to disk. I would appreciate it if you could enlighten us as to what is the problem.
--Antony Jankelowitz, (firm indeterminate)

Dear Adrian:

Can you please tell me where I can find indepth documentation for rule tool?
--Kenny Henderson, (firm indeterminate)

Dear Adrian:

We are using Solaris 2.3 on SPARC 20 (four 125-MHz CPUs). We have 15 to 20 idle time left over when we have 18 application processes running with lot of messages being processed. There are two processes which take up 35% and 25% of the time, respectively.

My question is: Is there anyway to change the time slice on Solaris? I don't know what the default is. But for example, if it is 20ms, we change it to 50 ms for the first process and 40ms for the second process. By doing this, there will be less swapping and better turn round. Do you think this will make any difference?

--Jay, (firm indeterminate)

Dear Adrian:

I'm trying to tune some SPARC machines. sar -w reports a lot of processes switches as reported in pswch/s. 1000 on a SS1000 60-MHz with two CPUs, more than 1500 on a SS1000 85-MHz with four CPUs.

1. Are these common values ?
2. Is pswch an accurate indicator ? Is it possible to reduce number of process switches without impact on throughput ?

Both machines are paging heavily (about 100 p/s). I used ioctls on /proc to get pr_usage_t's pr_minf and pr_majf (minor and major page fault). I have almost only major page faults.

I ran se2.4 on both systems and it reported nothing wrong. They have quite different profiles : One is an NFS server, one runs six Oracle instances (DB server).
--Alexis Grandemange, (firm indeterminate)

Q:Is pswch an accurate indicator?
A:Yes, the metric itself is accurate, but it is not usually useful as a problem indicator.

Q:Is it possible to reduce number of process switches without impact on throughput?
A:No, usually as throughput increases so do pswch. Use mpstat to see how many are involuntary versus voluntary context switches. If there are a large proportion of icsw, then increasing the timeslice might help a little. (see dispadmin) Don't expect any dramatic improvements.

% mpstat 5 CPU minf mjf xcal intr ithr csw icsw migr smtx srw syscl usr sys wt idl 0 68 19 0 1046 826 568 57 0 27 0 285 7 13 0 80

here there are 568 switches but only 57 are involuntary, Q:Both machines are paging heavily (about 100 p/s).
A:That's not very heavy. Its probably just filesystem I/O activity. Q:I ran se2.4 on both systems and it reported nothing wrong.
A:In that case they are probably both OK. Wait until a problem is reported before you start to worry.

Dear Adrian:

I would like to add a few points to your comparison of static vs. dynamic linking:

1. If most of your programs are small and contain mostly library calls, you are probably right that dynamic linking is faster and uses overall less memory. However, if your typical program contains mostly data, such as in FEM or CFD analysis, the memory saved by dynamic linking is negligible. On the other hand, in those applications, static linking provides for better linker optimization and (slightly to considerably) faster code.

2. Dynamic linking is fine so long as you have all necessary libraries installed on every computer where your program is to run. Depending on the nature of your libraries, this may be expensive, cumbersome, or impossible. Expensive, because you may have to buy an extra license of a compiler for every computer, instead of just using the compiler on one computer and then using the executable on any other computer.

Cumbersome, because you may have to crossmount libraries on a slow network, or just put a separate copy on each computer. Again, the alternative is to use just one computer for compilation and then take the executable elsewhere. Impossible, because you cannot put each and every runtime library on every computer, and not always is there a network available -- just think of a laptop with limited disk space away from the office.

In summary, dynamic linking may be better if you do not mind paying good money for license fees, disks and networks. It is definitely better if you are the person selling those licenses, disks and networks. Otherwise, better take a second look at static linking.
--Hubert Meitz, (firm indeterminate)

The benefit of dynamic linking is that you can upgrade to an improved libF77.so without rebuilding your application. You may also get a platform-specific libF77.so that is optimized for the hardware that it is running on. This may make a big difference for high-end floating point applications.

Dear Adrian:
I spent last week in Sun Server Admin class and the instructor mentioned that you have software that analyzes the Sun system based on your book. If so, I am very interested in obtaining this software. I believe it was called virtual adrian but I can't find it when doing an Internet search.

--Jim Hill, (firm indeterminate)

Dear Adrian:
Can you please mail me the ftp site for the Adrian performance tool ASAP! Thank you!!
--David Fuentes, (firm indeterminate)

Dear Adrian:
Following the guidelines indicated in appendix A of your SP&T book we have a four-processor SS1000 in a "Level Red" Mutex Stall (smtx > 400). Is there a quick hardware fix for this problem (i.e., more or faster CPUs)? Any advice would be appreciated. ...BTW, great book!
--Todd Resnick, Duke University Medical Center

See the SE Performance Toolkit Version 2.5 for more information.

If you have problems with this magazine, contact webmaster@sunworld.com

URL: http://www.sunworld.com/swol-03-1996/swol-03-letters.html
Last updated: 1 March 1996

If you have technical problems with this magazine, contact webmaster@sunworld.com

URL: http://www.sunworld.com/swol-03-1996/swol-03-letters.html
Last modified: