lkml.org 
[lkml]   [1996]   [Dec]   [12]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: Sybase for Linux?
From
Date


this thread has been going on forever in the sybase newsgroup -- i
thought i'd bounce it off the kernel folks since scott keeps referring
to how the kernel lacks certain features required for 'decent
performance' --


From: gray@voicenet.com (Scott Gray)
Subject: Re: Sybase for Linux?
[...HEADER...]

In article <slrn5abcdu.f58.ise118@vtaix.cc.vt.edu>,
Ram Kalapatapu <ise118@vtaix.cc.vt.edu> wrote:
>
>Does anybody know of an important contact at Sybase I can talk to about
>a port of Sybase SQL Server to Linux? I think Sybase is loosing an
>important business oppurtunity by ignoring the second most popular
>flavor of UNIX. I would also like to hear from fellow netters about their
>views on Linux as a dataserver platform especially in light of the
>significant changes in 2.x version of the kernel.

Unfortunately, from my understanding of the recent Linux kernel, there
are several features missing in order to make Linux a viable platform
for SQL Server, the largest being the lack of async i/o, raw partition
devices, and locking shared memory. None of these are required in order
to get SQL Server to run, however they are required to get anything
resembling descent performance and reliability on the platform.

Aside from these problems, there is also the fact that adding a new
supported platform to SQL Server is a very expensive proposition for
Sybase. It requires training of technical support people, setting
up a porting group and a sustaining group, marketing, etc. The long
and short is that I would never expect SQL Server to offered as a
supported product.

As an unsupported product, it would be possible. Most people don't
expect blazing speed or top-of-the line performance from "freeware"
so the feature issues would be reduced, and there is no training
issue, and a marginal cost of paying someone to do the port (although
I know several people in engineering that would offer to do the port
on their own time if Sybase would let them). But...I wouldn't really
hold my breath for this...Sybase has already had a bad experience
with releasing ct-lib to the Linux community (apparently they were
fielding support calls to Linux users which were explicitly told
that the port was unsupported). Also, SQL Server is their flag-ship
product (well, under the name "Sybase" and not "PowerSoft"), so I
find it highly unlikely that they would give out a free or
unsupported version.

-scott

P.S. I have no direct connection to Sybase engineering so these are
all my opinions.

--
Scott C. Gray gray@voicenet.com
Sybase Professional Services scott.gray@sybase.com
http://www.voicenet.com/~gray/sqsh.html

[SNIPPING SEVERAL POSTS]

gray@voicenet.com (Scott Gray) writes:


In article <58p6ad$aeg@dismay.ucs.indiana.edu>,
Brian Wheeler <bdwheele@indiana.edu> wrote:

> I really would like to see server available!
> Brian Wheeler
> bdwheele@indiana.edu

I applogize for re-ordering your post...I just wanted to point out
what may not be obvious...I very much agree with your sentiments. I
would very much like to see SQL Server available on Linux, but all
that I have been saying is that I can understand why it hasn't
been done yet.

However, I really don't see why there isn't a (new) port of ct-lib,
db-lib, and open server for Linux. this is a relatively low-cost,
and low-maintenance solution. This is what I have been pushing
for internally (to what little extent I can).


>In article <58n3nu$h4p@omni1.voicenet.com>,
> gray@voicenet.com (Scott Gray) writes:
>
>>Well, good luck getting 100 people to caugh up $5-20K for a product that
>>simply will not keep up with other platforms. I think that financially
>>it would be unsound to have SQL Server ported to LInux right now,
>>however I would very much like to see Open Client ported...I think Linux
>>makes a great client platform.
>
> Why will it not keep up with other platforms? That's up to Sybase,
>and no reason why it couldn't keep up at all.

As I have said, right now Linux lacks some of the necessary kernel
features for Sybase to perform well. I mean, what Linux has now
it does _very_ well, but it still needs some work in order to compete
(at least in terms of SQL Server performance). If I had the kernel
internals knowledge, I would attempt to tack some of this myself.

After reading your post and some of the others, I must say it would
be an interesting idea to provide SQL Server at a reduced price for
Linux to make up for the performance differences...this would be
very attractive to someone like myself that simply wants a server
to talk to while I do development, simply to make sure that my code
works. Unfortunately, Sybase's marketing is simply not geared
towards this sort of environment...however Caldera's is...hmmm.

> As far as 100 people go, I could easily see my university wanting a
>port of the server. That's worth some money, I'm sure. With the license we've
>got now Sybase software comes piled (not stacked) in bigish boxes. I'm sure
>the $5-20K figure would be a small percent of our current license.
> Since Linux is being used quite heavily in universities, having a
>major database like Sybase available would most likely a) let students get
>familiar with it while they're students and b) influence those students to
>use it instead of another DB product later in thier careers.

I strongly agree with this, and I am happy to say that there is at
least a movement within Sybase to provide cheap/free SQL Server to
various universities exactly for the reasons that you provide.

> Why do you think it would be financially unsound? Sure, Sybase would

As has been pointed out, it is a very expensive ordeal to support
a new platform. Between training/hiring new tech support people, hiring
new porting and sustaining groups, setting up a new regression testing
environment, marketing, etc. it ain't cheap; the expense is not in the
physical port, but in the "aftermath" of the port.

> If you're to listen to MS Reps, Sybase is already financially unsound
>and will go out of business any day now. I don't believe that, because MS
>Reps tend to be weasels of incredible degree. However, I think taking this
>financial risk would pan out in a good way.

Stonebraker has a great quote in "Readings in Database Systems"
concerning Oracle (although I don't think it is his quote):

Q: On what platform does Oracle perform the best?
A: A 35mm slide projector

Yeah. Oracle marketing tends to be a little overagressive.

> Someone mentioned (either here or in one of the Linux groups) the
>problems with releasing software for Linux due to the differences in
>distributions. As a solution for this, releasing a version and specifying that
>its supported under a single distribution. By also releasing what libraries
>(and versions) the product is linked against, people on the net will be able
>to "port" it to the other distributions.

This is essentially what Caldera does. You can only have support
for WABI if you are running on a pretty much virgin Caldera distribution
(i.e. you havne't upgraded the libraries, the kernel, etc.). Unfortunately,
a lot of people (such as myself), get a little frustrated when they
are a year or more behind the current release of the OS. Caldera
has a very good reason for doing what they do, but it doesn't fit
into my style of doing things. I'm still going to buy WABI, but
I am crossing my fingers that I don't need technical support.

> If nothing else, the newsgroup has gotten more lively since this
>whole thread started, so there is a good amount of interest in the product.

Like I said, I would dearly *love* to have SQL Server as well (I started
trying to port System 10 myself, but I just didn't have the time),
but I think it is going to be a long hard battle to convince the
powers-that-be within Sybase to support such a prospect; I mean, they
already feel burned by the CT-Lib on Linux ordeal.

Hopefully, if we (Sybase employees) can convince someone to at least
re-release CT-Lib on Linux that will be a first step in the right
direction (wish us luck!).

> Also, Linux is substantially faster than NT on the same hardware, so
>even if there is a performance hit in the linux version, it should be as fast
>as NT.

At the moment, I don't think this would be true. Until async i/o
is part of the kernel, NT will probably win. But, I certainly can't
wait for the day that Linux stomps NT in the database server
category :)

-scott

These are strictly my opinions, and not those of my employer.

--
Scott C. Gray gray@voicenet.com "my keybard is brken"
Sybase Professional Services scott.gray@sybase.com
http://www.voicenet.com/~gray/sqsh.html

\
 
 \ /
  Last update: 2005-03-22 13:38    [W:0.026 / U:0.408 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site