Sybase NNTP forums - End Of Life (EOL)

The NNTP forums from Sybase - forums.sybase.com - are now closed.

All new questions should be directed to the appropriate forum at the SAP Community Network (SCN).

Individual products have links to the respective forums on SCN, or you can go to SCN and search for your product in the search box (upper right corner) to find your specific developer center.

ASE on LINUX - the future ??

4 posts in ,  LinuxProduct Futures Discussion General Discussion Last posting was on 2002-03-16 17:53:42.0Z
Russ Wheaton Posted on 2002-03-13 13:57:24.0Z
From: "Russ Wheaton" <russman@wheatonworld.com>
Subject: ASE on LINUX - the future ??
Date: Wed, 13 Mar 2002 08:57:24 -0500
Lines: 25
X-Priority: 3
X-MSMail-Priority: Normal
X-Newsreader: Microsoft Outlook Express 6.00.2600.0000
X-MimeOLE: Produced By Microsoft MimeOLE V6.00.2600.0000
Message-ID: <jk9CCgpyBHA.318@forums.sybase.com>
Newsgroups: sybase.public.ase.general,sybase.public.ase.linux,sybase.public.ase.product_futures_discussion
NNTP-Posting-Host: 199.82.28.103
Path: forums-1-dub!forums-master.sybase.com!forums.sybase.com
Xref: forums-1-dub sybase.public.ase.general:3038 sybase.public.ase.linux:300 sybase.public.ase.product_futures_discussion:139
Article PK: 51345

Companies are inevitably looking at more cost effective HW solutions - AKA
Intel based LINUX serves.

If I am in the traditional land of big business data servers, I am used to
dealing with 10+CPU sun / hp machines for my more heavy workload database
apps.

Now switch me to LINUX - 2CPU based Intel machines.

Most likely I just need to be directed to the information, but here is my
question:

Without adding a routing layer to my application (or other costly things
that give my programmers bad dreams), is Sybase (ASE) keeping up with the
industry desire to run the same db app on FIVE 2way LINUX servers vs ONE
10way sun/hp box?

I think Sybase had better be. If they're not, I have some ideas. If they
are I want to review the white paper / product doc / engineering note / bar
napkin that describes the proposed architecture.

Russ Wheaton


Sethu Posted on 2002-03-16 17:53:42.0Z
Message-ID: <3C9386A6.223F644F@sybase.com>
Date: Sat, 16 Mar 2002 09:53:42 -0800
From: Sethu <sethu@sybase.com>
Organization: Sybase, Inc.
X-Mailer: Mozilla 4.79 [en] (Windows NT 5.0; U)
X-Accept-Language: en
MIME-Version: 1.0
To: Russ Wheaton <russman@wheatonworld.com>
CC: Sethu Meenakshisundara <sethu@sybase.com>
Subject: Re: ASE on LINUX - the future ??
References: <jk9CCgpyBHA.318@forums.sybase.com>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Newsgroups: sybase.public.ase.general,sybase.public.ase.linux,sybase.public.ase.product_futures_discussion
Lines: 247
NNTP-Posting-Host: 10.22.91.118
Path: forums-1-dub!forums-master.sybase.com!forums.sybase.com
Xref: forums-1-dub sybase.public.ase.general:3036 sybase.public.ase.linux:298 sybase.public.ase.product_futures_discussion:137
Article PK: 51343

First off, In our 3 year roadmap, we have plans to support Linux blades.

There was limited discussion in the sybase linux isug email alias list.
This is what I answered and I'm also attaching some of the following emails
that came thro'.
-------------------------------------------------------------------------------------
subject: Re: Using clustering with Sybase
Author: Sethu <sethu@sybase.com>
Organization: Sybase, Inc.
Date: Thu, 29 Nov 2001 19:51:18 -0800

First off, currently, ASE cannot be used in a clustered environment for
scalability. However it can be used for Availability. However, I want to
add some of my comments on using Clusters for Scalability.

A disclaimer before I start.. This is IMHO.

I strongly believe clusters are good for read-only situations.
They are not at all good for OLTP.

1) If you consider shared disk, the locking contention (which Oracle
claim that they have solved it), will soon show up within 3-4 nodes.

Oracle did a benchmark with Compaq Trucluster GS320 (32-way)
8-Quad system that showed only 5 percent better performance
number (officially published TPCC number) than a 24-way
IBM S85 number.

Compared to the shared disk, we have seen true SMP (Starfire,
superdome) have given 10-30 times scaling comparing
a single CPU system and the largest system on the same hardware.

2) The GS320 along with the interconnect cost ton of money. Plus you
need multi-ported disks. As far as I know these multi-ported disks
cannot scale more than 4 nodes.

3) If you consider shared nothing (which we will eventually implement in ASE),

unless the user does some kind of load balancing, the function shipping
to the other node and doing the update (for short burst transactions)
far overweigh the scalability improvements one will get.

4) Analysts predict that only 10% of Oracle community will
use ORAC by year 2006. (that is 4 years from now).

My philosophy is

1) If possible do horizontal partitioning and configure it for
availability
cluster pair. You'll get better scalability and better availability
for OLTP
applications.

2) If horizontal partitioning is not possible do vertical partitioning
(more
CPU, faster CPU, more memory all will help to scale)

Systems like Beowulf are very good doing heavy-duty read-only number crunching

jobs. The challenge using Beowulf for databases is that, database typically
involve lots of I/O. Anytime you do I/O to a single client, the client
eventually
becomes bottleneck. Some solution there would be make our OpenClient
cognizant of database schema and clustered nodes and fire off queries
in parallel using parallel connections at the client side. This is what
MPP did. The drawback with MPP was that knowledge of client side caching
and client side parallelism was not in the OpenClient but was done in
the ControlServer (OpenServer based).

As I said this is MHO.

Thanks,
Sethu

------------------------------------------------------------------------

Subject: RE: Using clustering with Sybase
Author: Nick Barbalich <nick.barbalich@medicals.co.nz>
Date: Tue, 4 Dec 2001 12:56:52 +1300

Dear Sethu

Could you clarify the following terms in your reply:
1. horizontal partitioning
2. vertical partitioning
3 ORAC

thanks
Nick Barbalich
--------------------------------------------------------------------------

Subject: Re: Using clustering with Sybase
Author: Sethu <sethu@sybase.com>
Organization: Sybase, Inc.
Date: Tue, 04 Dec 2001 15:15:53 -0800

Horizontal Partitioning (HP)

HP can be done thro'
(a) Appl/DB cognizant and
(b) clusters

(a) Appl/DB cognizant

- Partition the application (application is cognizant
of the partitions) either with geographic boundaries,
some other range values meaningful to the application
that is being partitioned.

(b) clusters

- done thro' clusters with application not aware of partitions.
This will not scale for some applications. Imagine that
the order number is generated. Now all the clients in all
nodes will try to get the same lock on the order table --
even if it is short duration and we get the lock, the amt
of msg being passed to maintain cache coherency will be
high.. According to Amdahl's law, if 1% of problem fails to
parallelize, then no matter how much parallelism, clustering
is available for the rest, the problem can never be solved

That is why Oracle has recommended the following page 63
of their P & T Guide for RAC that says.

"To reduce Real Application Clusters overhead, each instance in a
cluster should ideally perform most DML operations against a set of
database tables that is not frequently modified by other instances."

Page 63
Oracle9 i Real Application Clusters
Deployment and Performance
Release 1 (9.0.1)
July 2001
Part No. A89870-02
http://download-west.oracle.com/otndoc/oracle9i/901_doc/rac.901/a89870.pdf


Vertical partitioning
- Single machine scaling up thro' # CPUs, memory, disk etc.
This is like going from 8 way-> 16 way -> 32 way on the same
box or thro' NUMA box

ORAC is Oracle Real Application Clusters

Sethu
PS: I never got any feedback on this topic. I would like to
get some feedback on what others think in this
email alias.

Thanks,
Sethu
-----------------------------------------------------------------------
Subject: RE: Using clustering with Sybase
Date: Wed, 5 Dec 2001 12:32:38 +1300

% 4) Analysts predict that only 10% of Oracle community will
% use ORAC by year 2006. (that is 4 years from now).

Sorry to single out this point when the whole point of your post was
something else entirely, but can you give some references on this? I'd find
it very interesting reading, personally.

-------------------------------------------------------------------------
Subject: Re: Using clustering with Sybase
Author: Sethu <sethu@sybase.com>
Organization: Sybase, Inc.
Date: Tue, 04 Dec 2001 16:02:50 -0800

Look at the following URL

http://www.informationweek.com/story/IWK20011121S0021

Sethu
----------------------------------------------------------------------------
Subject: RE: Using clustering with Sybase
Author: Nick Barbalich <nick.barbalich@medicals.co.nz>
Date: Thu, 6 Dec 2001 15:59:50 +1300

Dear Sethu

So in theory, you may be able to "port" a application which runs on a large
32 cpu server to run on (say) 8 separate-but-linked 4 way CPU boxes as long
as the application is cognizant of the partitions and you use ASE 12.5
Distributed Transactions Management Features to coordinate the transactions.


I assume this would work BUT,

1. you would have to manage 8 servers verses one 32 cpu server.
2. writing the code would be more involved
3. management of recovery and synching the time on eight servers would
be more difficult.
4. the trade-off is "ease of management" verses "lower hardware costs".


Nick Barbalich
-------------------------------------------------------------------------------

Subject: Re: Using clustering with Sybase
Author: Sethu <sethu@sybase.com>
Organization: Sybase, Inc.
Date: Thu, 06 Dec 2001 05:31:32 -0800

Hi Nick,

I agree with your points. However there are several trade-offs.

If you want high scalability you have to go with the horizontal
partition. There the trade-off is between "throughput/performance"
and "ease of mgmt" and "lower hardware costs".

Sybase systems are easier to manage than Oracle systems. Sybase
can do better with managing systems thro' wizards/gui etc.

Regarding point #3, if your system is horizontally partitioned say
by Service Level Agreements (Gold Card Members vs. Platinum Card Members vs.
Titanium Card members). The following are the advantages in this setup.

1) You can setup HA only for Titanium card members
(as a privilege to the high end credit card user )
You can use the Gold Card members server as the HA for Tit. card members

2) If gold card member server is down, only the gold card members services
are down.

On the contrary if you are using shared disk clusters, you need a real fast
interconnect -- otherwise the interconnect can become the bottleneck.
To get good perf. you need myrinet kind of interconnect with shared disk.

Cheers,
sethu
---------------------------------------------------------------------------
Subject: RE: Using clustering with Sybase
Author: Michael Peppler <mpeppler@peppler.org>
Date: Thu, 6 Dec 2001 09:20:16 -0800

One way I've seen it done is via an ad-hoc OpenServer that serves as
both a switching and a connection pool system, and which directs
requests based on some of the parameters.

Another way to do this is to separate your data into vertical and
horizontal data. The vertical data needs can be easily partitioned
(i.e. one row in the vertical data section doesn't need access to rows
in the vertical data that aren't on this server) while the horizontal
data needs to be the same on all the boxes. You then use replication
to keep the horizontal data in sync.

It's non trivial, of course, but should be fairly scalable. For a
large web site, for example, it would also have the advantage of only
requiring shutting down service for a subsection of the site if one of
the data servers is down (maintenance, etc.)

Michael

---------------------------------------------------------------------------------


Lindsey White <whitelw Posted on 2002-03-13 15:56:32.0Z
Date: Wed, 13 Mar 2002 10:56:32 -0500
From: Lindsey White <whitelw@*NOSPAM*michigan.gov>
X-Mailer: Mozilla 4.72 [en]C-CCK-MCD (WinNT; U)
X-Accept-Language: en
MIME-Version: 1.0
Subject: Re: ASE on LINUX - the future ??
References: <jk9CCgpyBHA.318@forums.sybase.com>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Message-ID: <rtkQpiqyBHA.204@forums.sybase.com>
Newsgroups: sybase.public.ase.general,sybase.public.ase.linux,sybase.public.ase.product_futures_discussion
Lines: 31
NNTP-Posting-Host: 204.22.154.244
Path: forums-1-dub!forums-master.sybase.com!forums.sybase.com
Xref: forums-1-dub sybase.public.ase.general:3037 sybase.public.ase.linux:299 sybase.public.ase.product_futures_discussion:138
Article PK: 51344

How about the multi-processor AMD solution that is supposed to be
available Q1 2003? I am clueless about Linux and it's ability to run on
4-8 cpu hardware. I assume that this hardware will cost less than
traditional SUN/HP/IBM etc... hardware.

Lindsey

Russ Wheaton wrote:
>
> Companies are inevitably looking at more cost effective HW solutions - AKA
> Intel based LINUX serves.
>
> If I am in the traditional land of big business data servers, I am used to
> dealing with 10+CPU sun / hp machines for my more heavy workload database
> apps.
>
> Now switch me to LINUX - 2CPU based Intel machines.
>
> Most likely I just need to be directed to the information, but here is my
> question:
>
> Without adding a routing layer to my application (or other costly things
> that give my programmers bad dreams), is Sybase (ASE) keeping up with the
> industry desire to run the same db app on FIVE 2way LINUX servers vs ONE
> 10way sun/hp box?
>
> I think Sybase had better be. If they're not, I have some ideas. If they
> are I want to review the white paper / product doc / engineering note / bar
> napkin that describes the proposed architecture.
>
> Russ Wheaton


nick.barbalich Posted on 2002-03-14 01:33:20.0Z
From: nick.barbalich@medicals.co.nz
Date: Wed, 13 Mar 2002 20:33:20 -0500
Newsgroups: sybase.public.ase.product_futures_discussion
Subject: Re: ASE on LINUX - the future ??
Message-ID: <455BD207CDA0639F00088B9F85256B7C.005C551785256B7B@webforums>
References: <jk9CCgpyBHA.318@forums.sybase.com> <rtkQpiqyBHA.204@forums.sybase.com>
Lines: 27
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-Transfer-Encoding: 7bit
Path: forums-1-dub!forums-master.sybase.com!forums.sybase.com!webforums.sybase.com!news
Xref: forums-1-dub sybase.public.ase.product_futures_discussion:668
Article PK: 94195

I think Sybase know that users want/desire a clustered version of ASE on
Linux.
But they have a couple of issues to deal with.
1. Bring out such a product, and they help screw up their relationship
with SUN/HP etc. As this move will rapidly increase the speed at which
people move off their middle to high end UNIX servers.
2. Linux verse SUN/HP is cheap. To quote as AOL anon employee, "AOL
number crunchers figure they can replace and $80,000 box running
proprietary UNIX with two $5,000 Linux boxes and get a 50% increase in
performance in addition to the cost savings." Source - Newsforge article
March 11, titled "AOL enbraces Linux & Mozilla"
3. Pricing. Would they still charge the same for ASE if it ran on 5 x 2
intel/linux servers verses 1 x 10 CPU Sun/Hp box. I'm sure this change
would hit their back pocket.
4. Do nothing option. I'm no Linux guru, but at present (version 2.4) it
does not scale above 4 cpu's. Linus has OK'ed a pretty radical rewrite of
the Linux scheduler by Ingo Molnar for 2.5.2-pre6, which has really
inproved SMP performance. Early indictions is that it runs well on a 8 CPU
box. [ Want more details go to:
http://www.uwsg.iu.edu/hypermail/linux/kernel/0201.0/0810.html ]

So if the users wait until version Linux version 2.6.0 is released in a
year or two, the hardware is more powerfull, the o/s will scale better,
they can keep their relationship with SUN/HP intact (by calling this
progress) and they will not of had to waste the money building a clustered
version.

The only question is what will Oracle / IBM have done in the mean time ?