mark@drd.com (Mark Lawrence) (04/24/91)
--begin included text Date: Mon, 22 Apr 91 23:54 GMT To: "mark@DRD.Com" <mark@DRD.Com> Subject: Reply to a remark Message-Id: <32910422235423/0003791961NB2EM@mcimail.com> We received the following comment via Internet on a story in our April issue: ...A point that some might find of significance in this benchmark is that the machine used was a dual CPU multiprocessor. The operating system OS/2 does not support symmetric multi-processing. The SCO UNIX does. Now the people at DBMS wonder why UNIX is faster... Apparently the person who wrote that remark missed the following part of the article (page 47): "SCO UNIX also has the advantage of symmetrical multiprocessing. LAN Manager 2.0's multiprocessing option allowed the LAN I/O and file system code to be dedicated to one processor while the rest of OS/2 and the database server ran on the other. Santa Cruz Operation's MPX option for UNIX 3.2.2 was not nearly so lopsided. While all IRQ handling would sit on processor 1 because of a SystemPro quirk, all other processes would be assigned to the less busy of the SystemPro's two 33 MHz 486 processors." We don't mind criticism. We just like it to be accurate. Kevin Strehlo DBMS Magazine --end encluded text -- mark@drd.com mark@jnoc.go.jp $B!J%^!<%/!&%i%l%s%9!K(B Nihil novum sub solem
gupta@cai.com (04/27/91)
In article <1991Apr24.152833.3214@drd.com>, mark@drd.com (Mark Lawrence) writes: > --begin included text > [...] > > We don't mind criticism. We just like it to be accurate. > > Kevin Strehlo > DBMS Magazine > --end encluded text > -- > mark@drd.com > mark@jnoc.go.jp $B!J%^!<%/!&%i%l%s%9!K(B Nihil novum sub solem The usefulness of a benchmark is that it points out a fact that may not be otherwise obvious. If we know beforehand that Operating system 1 (SCO UNIX) provides symmetric multiprocessing while Operating system 2 (OS/2) requires all database work and substantial OS work to be done by a single CPU (except for LAN I/O and file system code), and a dual-cpu machine is used for benchmarking, then the multi-user performance of the machine running Operating system 1 (UNIX) is likely to be twice as fast as Operating system 2 (OS/2). I leave it to the net to decide whether an elaborate benchmark is needed to prove this. However, it would be extremely useful to know which operating system is better on a single-cpu machine with 16MB of memory (if that is the maximum that OS/2 can support). This might give us insights into which operating is actually better. For example, what is the overhead of context switches? What is the I/O performance under these operating systems? What is the overhead of inter-process communication? Etc. Benchmarking is very hard (I know). It is very important that at its conclusion one is very clear about what the benchmark has proven. The DBMS magazine article confirms that if I had a dual-CPU server I should use SCO's MPX option for UNIX (an operating system that supports symmetric multi-processing) as opposed to OS/2. However, if I have a single-CPU server, I could not draw the same conclusions. Can you? Regards, Yogesh Gupta Disclaimer: The opinions expressed in this article are mine and do not represent the opinions of anyone else. If you agree with me ...