Letters to the editor

NFS info, anyone?

I am trying to write a research paper on distributed file systems and would like to find some info on NFS. Would you please tell me where I can find a ftp or a web site which has info on NFS? Thanks.

--Alice, (firm indeterminate)

Career Advisor column: NT or Unix?

Interesting article. Even I face the same question once in a while. I seem to be over depending on Unix. Also, don't you think Unix scores over NT because it is already ported to various architectures? I have not heard NT running on a non Intel platform.

--Vijayanth Devadhar, Graduate Student, Wichita State University

Whether people actually buy and use NT on Alpha and MIPS is another question...

Editors:
According to your article, it seems that Unix-based systems are being pushed back from the front-end to the back, because of NT's interface and application software library. You say that the reasons why Unix will remain is because it is reliable, powerful, mature, etc.

After reading your article, I get the feeling, what's the point of Unix then? Being regulated to the "back room" is not my idea of a career choice! Imagine saying to people, "Yeah, I work in the back room on our Unix machines." And what happens when NT becomes (if it becomes) as robust and dependable as Unix? Better yet, what happens when it becomes almost as robust and dependable? Then it seems to me that for the same reasons you outline as to why NT is taking Unix over in the front end, NT will kick Unix out of the back-end too!

Seems to me that the best career advice you would give to anyone is to forget Unix and jump on the NT band wagon (which it seems you have done with your PNT company). So why not just tell people to forget Unix and go directly to NT?

I would like to ask one other question, IBM will soon be releasing a microkernel based version of OS/2. Provided IBM gets its act together when it comes to marketing and support (which it seems to have improved on for OS/2 Warp), the benefits/advantages of NT over OS/2 will be reduced significantly. (Although some would argue that OS/2 is a better choice right now, I don't however, because I think reliability is one of the most important factors, otherwise I think OS/2 is a better OS for every day use.) So how do you think this new version of OS/2 will work out? OK, I know its a crystal ball question! ;-)

--Frank Fabbrocino, (firm indeterminate)

Editors:
I am in much the same position as the author of the letter that prompted Edgar Saadi's article on NT vs. Unix: almost 20 years experience with Unix, no experience with NT, and wondering whether NT will supplant Unix. So I was very interested in the article. Unfortunately, I was disappointed.

Although Saadi says that "Unix is here to stay", the only reasons he gives to back up this opinion are: (1) NT hasn't been around long enough to be trustworthy for large sites; (2) NT lacks some features of Unix, such as true multi-user operation (some things can only happen at the console) and top networking support. Time will cure these deficiencies, and at the rate NT development is going on, it won't take too long.

As far as I can tell, the only advantage Unix has over NT is its tradition of openness, including the free sharing of source code. This will probably mean that Unix will always be ahead of NT in leading-edge development, because of the leverage we get from working with and adapting other people's code. This does not mean that Unix will maintain a significant market share; it does not mean that Unix programming jobs will always be plentiful; and while I agree that "Unix is here to stay", so are JCL, OS/MVS, and Multics. I wouldn't recommend that anyone start a career in any of them today.

I agree wholeheartedly with Saadi's final comment that Unix programmers have an edge over DOS/Windows programmers in understanding the power of NT because it is the first OS from Microsoft that actually comes close to the power of (today's) Unix. So I'm learning about NT as fast as I can!

--Dan Franklin, (firm indeterminate)

The Career Advisor column's mission is not to weigh the technical merits of operating systems. Columnist Saadi attempted to answer the question (we paraphrase here), "I know Unix. Should I learn NT?" His answer, in summary: "Not a bad idea."

PC X servers compared

Thank you for your very nice and helpful article in SunWorld Online about PC X servers. Do you know of any articles by yourself or others which review and compare PC X servers connected over phone lines instead of Ethernet? You had some comments in this regard but it was not a major focus of your article. Since the serial link is likely to be a bottleneck, the performance results which you obtained with Ethernet connections may not correlate well with results for a X over a serial line.

--Bill Rose, Neural Computation Program, DuPont Company

Editors:

I enjoyed the article. However, would it possible to received a copy of your findings on the other products which you may have tested? That is, did you test software from any of the other vendors which you listed at the end of the article?

--Jacques Dejean, Unix Systems Analyst, Brown & Root, Inc.

Phillips reviewed the top PC X servers only, and does not hot have any test data on the less popular packages.

TCP/IP primer feedback

First let me complement you on your article.

I do have a question on where to find reference documents on how one goes about subnetting a routing domain. I am not referring to RFC 950 or the mechanics but rather taking a organization and looking at its locations and user communities and applications so as to put together a subnet plan. How many subnets are required, regardless of what address it might own. The appropriateness of VLSMs versus straight subnetting etc.

--Richard N. Lerner, Golden Gate University

Once you know your load, it's time to figure out how many clients you can put on a wire at one time. The loading you can tolerate on each wire will drive the number of subnets. for example, 30 NFS clients that each send about 5 NFS ops/second will use just about half of a 10 Mbit/sec Ethernet. If you have 100 clients, figure on three or four Ethernets to start.

The implementation of virtual subnets versus "straight" subnetting should be driven by implementation cost and scalability. Will you be able to add more subnets later? What is your cost per subnet? Will your subnetting strategy handle all of the types of traffic you want to route (including any non-IP traffic)?

Editors:

I would like to know on Thin-Net, Thick-Net, Twisted Pair or FDDI how many nodes can you have on a network before using devices such as a repeater, bridge, or concentrator what is the common practice for safe LAN load? And do different manufactures have different specifications for this?

--Phillip, (firm indeterminate)

The maximum number of nodes is a function of the maximum cable length permitted. Adding a repeater doesn't change the answer, because a repeater is a physical layer extender. A twisted pair concentrator also doesn't change the answer, because it is another type of physical layer device.

The maximum number of devices ranges from about 150 to upwards of 200, but you're likely to hit throughput limits long before you wire in the maximum number of devices. With 100 nodes on a 10 Mbit/second network, you're only going to average 10 Kbytes/sec per node. If your users are doing heavier work, such as sharing files via NFS or browsing the web, you'll want to put fewer machines on each network. See the answer to the previous question about capacity planning to get a better idea of determining practical limits

Editors:
Thanks for the TCP/IP primer! Do you know of a router primer (html or book) source?

--Kevin Sears, GE Medical Systems

Praise for golden gateway

Hello Hal, excellent article. It reminds me of a very large mainframe project we just finished rehosting to unix here at our L.A. Outsourcing Center. A number of CICS regions and Cobol code have been lifted and now run concurrently along with their mainframe counterparts. In fact, the more I read the article, the more I thought you were writing about this project. As expected, it is a very large rehosting project. And its success is leading to more projects. gains would be met from this task. The key issues you mention on gateways and re-engineering practices are exactly what is being implemented here.

I was wondering if you have any other reference material on rehosting applications from mainframe to Unix??

--Neil Greene, SHL Systemhouse

`Toy Story' powered by 117 SPARCstations

I would like to have seen a picture of the wall of 117 SPARCstation 20 servers.
--(name and firm indeterminate)

Reader questions to the performance columnist

You have not focused on 64-bit file systems and performance in your article. Will I/O be faster on a 64-bit file system and special on a database application like Oracle?
--(name and firm indeterminate)

Oracle runs best on a raw disk setup, and there are no 64-bit features that would speedup file system accesses.

Adrian:

We have a SS20/712 as an applications layer firewall that is (at times) completely CPU bound w/ all the http traffic going through it. We are in the process of enhancing the http-proxy w/ all the recommendations made on this web page. However, until that is done we want to increase the throughput via hardware, i.e., faster processor. We are looking at the 100-MHz HyperSPARC setup, but don't know what the optimal cache size would be. We have a choice of 1M or 256K. Please help. In a networked environment what would be the preferred (fastest) for us?

--Mike McPherson, (firm indeterminate)

I would spend the money on an UltraServer 1 Model 170

Solaris 2.5 is a bit more efficient than 2.4, and the faster CPU and system bandwidth will probably work better than a dual CPU SS20.

I don't think HyperSPARC systems run kernel code as well as SuperSPARC systems. I found the 125-MHz 256KB HyperSPARCs were about the same as 60-MHz SuperSPARCs for running commercial applications like database backends that do a lot of kernel work.

Is the http proxy forking for every request? If so, a preforked or threaded proxy would be much better -- i.e., Netscape or phttpd or Apache, but not CERN or NCSA.

Adrian:
I attended a seminar you gave at a Computer Literacy in San Jose a while back and remember you mentioning a caveat about using CacheFS.

I remember you saying something like "it's not a good idea to use CacheFS on a r/w filesystem." What I can't remember is WHY.

Is it because writes through CacheFS are slower, or is it because writes through CacheFS are unreliable? Or does having an r/w fs mounted through CacheFS cause performance of CacheFS to drop in general?

Also do you have any suggestions for CFS option settings for read-mostly filesystems ?
--Jim Burwell, Systems/Network Admin., Broadvision

Read-mostly is fine. If you only read the data once don't bother caching it, if you keep changing a lot of it it is a waste of time caching it.

If you have a few updates, but mostly read the data it should give a good speedup. /var/mail is a really bad choice, /home is usually OK, /export/local (or whatever you mount applications on) is a good idea.

Great Java questions

Will Java become the platform-independent language standard of the internet?

1. Why is platform-independent code important? Shouldn't we merely develop a protocol which uses OLE or OpenDOC enabled applications and have a means of downloading the appropriate versions of the software as they are needed?

2. Why use C++? The principle of platform-independent code is ruined when you chain a developer to a language that was intended for developing Unix terminal-based applications. (I have noticed that Java happens to be an excellent tool for programming terminal-style applications.) The success of the Web hinged upon the ease of which HTML was coded and parsed...does Java offer us this?

3. What about standards? Can I create a 3-D visual programming language which compiles to Java's byte-code?

4. Libraries? Who is deciding what libraries are important to distribute with the Java interpreter? Is there a 3D modeling library? Is there a serial communications library? Joystick functions? Can I develop a proprietary library compiled for use on multiple platforms and integrate it seamlessly into Java?

--(name and firm indeterminate)

Rave for RAID story

In an effort to make an intelligent purchase decision for our digital video lab here at school, I read your article on RAIDs with much interest. We are attempting to expand our lab capability so students can digitize footage and then use our editing software (Premiere and Soft F/X) to learn non-linear editing. We presently have five Mac platforms each with it's own 1 gig drive and connected through a standard Ethernet LAN.

Unfortunately, the gig hard drives don't provide enough capacity to provide storage for projects and other software, so we have decided that we may need a server and RAID. We've concluded that a Apple Power PC 9500 would be a good choice for the server, but we're not sure what sort of RAID will best suit this application. Since we're a school, we have to make the most of the money we spend, so I am hoping we can come up with a less expensive, but effective solution.

Would you be willing to suggest a system (brand, model, RAID level, etc.) that would be cost effective and serviceable? I'm assuming that one large RAID can service a number of workstations -- we hope to expand to a total of 15. Also, what sort of network enhancements would be necessary (fast and wide SCSI, etc.) would be required to support the transfer of data from the server to the workstations?

I realize you must have many important things to do besides answering all the e-mails you must have generated from your article, but we haven't been able to find anyone who has the expertise to answer our questions. Thanks in advance for any light you can shed on our ignorance!

--Ron Bellamy, Program Coordinator, Video Production Technology, Pellissippi State Technical Community College

Editors:
Very good article, covered many areas of impact which I had not been directly involved with, however one small comment, although you say that writes to the mirror/parity disc may only take 15-20 milliseconds extra, what we've found in real life is that some of our clients who have RAID-5 subsystems notice this and obviously regard it as a performance problem.

The importance of battery-backed NVRAM caches in this instance cannot be stressed enough, is it possible to have NVRAM cache on other RAID? devices.

--Rob, (firm indeterminate)

Regarding the second reader's letter, Brian responds, "I don't think I said `writes to the mirror/parity disc may only take 15-20 milliseconds extra.' In practice, we find that writes to mirror disks are about 15-20 percent slower than a regular disk. Usually this is a lot less than 15-20 milliseconds. Usually it's on the order of 1-2 ms additional delay (out of 5-12). Most users don't notice this too much unless of course they're on the performance edge. Mirroring is pretty reasonable for writes, which is one reason so many people use it.

"RAID-5 is a different story. In RAID-5 you can be subject to writing the disk twice within 15-20 ms, so this is really 30-40ms, and that's a lot slower than a regular disk. Clearly this is a performance problem.

"Most RAID devices permit battery-backed NVRAM caches. The SPARCstorage Array certainly does, as do most competing high-end RAIDs."

Client/Server column questions

With great interest I have read your nicely worked out article "The Open Document Management API" in October's SunWorld Online. The article inspired me to find out more about the ODMA, but to my great astonishment, "my" common WWW-search-engines did not deliver so much more information.

My question is, whether the ODMA-standard-draft is available publically anywhere in the net. Do you have any information about it? Could you give me a little hint ?

--Wolfgang Schulze, Research Scientist, Database Group Dresden University of Technology

Mr. Rosenblatt:

I have just read your excellent article in SunWorld Online entitled "The Open Document Management API."

I have as yet been unable to find any leads on where I might get to read the ODMA Specification or even follow its progress.

I would be very grateful if you might find the time to return to me a URL or address that I might further my investigations into the most interesting subject.

Sorry for the intrusion but I really have been unable to come up with anything using Lycos, Yahoo, etc, etc.

--Nick Biggs, Principle Analyst Programmer, Glaxo Wellcome

PS. Should you find time to reply I will be sure to post any information you can provide to the Usenet group comp.text where I have seen a number of other similar enquiries to this one of late.

My year with Scott

Excellent article. Well written. Proper balance between being informative and entertaining. Article avoided writing an article that just told readers how "wonderful" the boss is.
--(name and firm indeterminate)

Editors:
Great article! I gained further insight into Scott and Sun. The article helped me to relate better to Sun as an enterprise of individuals learning, growing and contributing. I like relating to enterprises as groups of people. The camera/film analogy was particularly insightful.
--(name and firm indeterminate)

Editors:
I just read the article about Barbara, who spent one year with Scott. It was a very interesting article. Did Scott ever work for anyone else than Sun (himself)? Did you (Barbara and Scott) have heavy arguing about some issues? What is the expected development in the computer SBU (trends, new products, Etc.) ?
--(name and firm indeterminate)

Scott and I did debate various issues. He kindly allowed me to freely question his decisions and vision for the company. I think he found it helpful to get a sense of how others perceive various actions as well as knew it was a great training experience for me.

We have all kinds of cool things going on in the computer operating company -- SMCC. We've just announced the UltraComputing line of desk top systems, we have a $1 million Java Cup International programming contest going on, and many exciting things for the next six months. Stay tuned, Sun has a lot to offer.

Job well done

Have been following your mag since sometime now, and have always found it worthwhile reading. The articles are very relevant, well written and deal with very "current" topics. The presentation of all this is pretty nifty too.

Keep the show going! My congratulations and best wishes to all of you at SunWorld Online. Thanks for giving one, real neat site !

--Raj, (firm indeterminate)

From our reader support dept.

Editors:

I used to save some of the back issues of the paper version of your magazine. Are the back issues of the electronic version available?

--(name and firm indeterminate)

To get free e-mail notifications of significant updates to SunWorld Online go to http://www.sunworld.com/common/swol-subscribe.html and complete the form.

All back issues of SunWorld Online are available. Go to the "Back Issues" selection on the table of contents.

Editors:

How about a cumulative product index that covers all issues? Thanks.
--(name and firm indeterminate)

Editors:

I've greatly enjoyed your new online incarnation of SunWorld Online. However, it would be very useful to see a searchable word index that allowed the reader to search the current issue and back issues for information about a specific topic. Apologies if I have simply overlooked the existence of such an index...
--(name and firm indeterminate)

An index tool for SunWorld Online is in the works. In the meantime, you can use the search engine at http://www.sun.com/ to search all of Sun's site, including SunWorld Online.

Editors:

You should support Adobe Acrobat's pdf encoding for articles in your on-line SunWorld. That way, when I want to save an article with nice graphics -- like Adrian Cockcroft's excellent article, Performance Tuning, in the Oct. edition -- I can later open it up under Acrobat and still see all the nice graphics. I can even print it out when I so choose.

--Dave Ford, (firm indeterminate)

Our stories stay available with their unique file names, so you may want to concoct your own personal mini-Yahoo of valuable URLs.

Thanks for your message.

Editors:
I'd like to see more (selections) available to rating an article than just "excellent," "good," and "poor." I think five levels would be better. This could also be done with the article length feedback.

Thanks for the good work. I like having the ability to quickly and easily go back to old articles and see what was said. One thing that I'd like the Web people to make sure they do is to make the link names consistent to a particular article from different pages. I use the web browsers ability of showing which links have been visited to keep track of those articles that I've read and those I haven't. Some times in the past, articles that I've read have been pointed to differently on a different page.

--Blair Zajac, (firm indeterminate)

Since we recognize many readers will want to bookmark a story for later reading, all stories have unique file names. This also makes it possible, as you suggest, for you to see if you've selected a story before since its hyperlink will indicate your prior use.

Editors:
I need written articles. Looking through a computer display can be very irritating. Please send all information about Java to the address below.

--(Name and firm deleted to protect the clue-impaired)

Netscape, Mosaic, and Lynx allow users to save pages to their local disks. In Netscape, go to the File menu and select Save as... Name the file, tell Netscape if you want to save the file as PostScript, text, or HTML, and press the OK button.

To print a story, go to the File menu and select Print...

All browsers offer many other useful commands. You should read the documentation for more information.

Those bothersome ads

I understand the need to place adds into these pages but I find it extremely annoying that they are placed in the body of the document. My comment is put them at the beginning or the end but please not in the middle of the text.

Thanks for listening...

--Kirk, (firm indeterminate)

Thank you for your note.

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

URL: http://www.sunworld.com/swol-12-1995/swol-12-letters.html
Last updated: 15 December 1995

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

URL: http://www.sunworld.com/swol-12-1995/swol-12-letters.html
Last modified: