Sybase IQ Experiences
Diederik Dellaert
I am working on a data warehousing project where we consider using
sybase adaptive server IQ for our datamart solutions.
I am looking for feedback of anyone who has used the product under NT
or any other platform.
Issues which I am interested in are (among others):
- Scalability vs. platform issues (what about NT?)
- Memory requirements
- Performance issues
and any other issues you can think of.
Thanks a lot for any information,
Diederik Dellaert
Joerg Menker
(Diederik Dellaert) wrote:
>Hi all,
>
>I am working on a data warehousing project where we consider using
>sybase adaptive server IQ for our datamart solutions.
>I am looking for feedback of anyone who has used the product under NT
>or any other platform.
>Issues which I am interested in are (among others):
>- Scalability vs. platform issues (what about NT?)
>- Memory requirements
>- Performance issues
> and any other issues you can think of.
>
Hi Diederik,
I'm working on a data warehousing project in Switzerland and we are
using Sybase IQ und Sun Solaris.
I don't know the performance of IQ under NT but NT is definetely not
as scalable as any UNIX and IQ loves memory and lots of disks. We're
using IQ on a ES6000 box with 8 cpus, 8 gigabyte of RAM and several
hundred gigabytes of disk space but still we would love to have more
of all. IQ scales great with more cpus and memory. It's also a good
idea to partition your indexspaces in order to reduce I/O.
Depending on the amount of data in your data marts (probably very
much) some gigabytes of RAM would be very useful (at least 1 gig). IQ
is great in utilizing memory.
The overall performance of IQ is great. Often it's several hundred
times faster than ASE. You've got to keep in mind though that you
cannot use IQ alone. Yopu've always got to have an underlying database
such as Adaptive Server Enterprise, Oracle etc.
IQ also has some limitations with regard to sql (not fully ansi
comliant) but maybe this has changed with the newest version (11.5).
We're using version 11.2 right now.
It's probably the best database for data warhousing with regard to
performance, especially when you're using ROLAP tools whose
performance is always poor.
Joerg
+--------------------------------------------+
| Joerg Menker men...@uni-muenster.de |
+--------------------------------------------+
| j.me...@saphir.net |
+--------------------------------------------+
RABK
We have been using Sybase IQ 11.5 for NT 4 SP3 on Compaq 6000R with 2 GIG
RAM with 60 GIG RAW DATA which equates to around 20 GIG in Sybase IQ. The
experience has been great so far. The longest query ever has taken 4
minutes. Typical queries return in < 20 seconds with 22 million records +
using 3-5 table joins. Data transformation and denormalization from the
various sources was a battle and the most timely portion of this project.
Sybase IQ has proven to be very fast, but lacks full ANSI-SQL compliant
standards, hogs memory, and loses memory occassionaly in NT, albeit thats
an NT alloc,free enhancement coming in NT 5.0.
Sybase IQ 11.5 for Sun Solaris 2.6 on (64 bit) provided us with 4 GIG RAM
SUN ES6000 16-CPU timely result for MOST queries. Data volume was 300 GIG
RAW compressed down to < 100 GIG in Sybase IQ. Battles with query tools
and data transfer and conversion were issues (COBOL VSAM files). The two
largest tables 150 million and 70 million used in joins in certain types of
queries caused long query times > 20 minutes. Overall, the query speed has
been satisfactory on this larger scale solution as well.
My previous experience with database vendors and the types of queries used
in DSS have not resulted in such low storage, high speed, and ad hoc
flexibility as Sybase IQ provides. Any data warehouse projects should
consider Sybase IQ to be one the leaders in these types of ROLAP database
solutions.
RABK
Diederik Dellaert <diederik...@be.pwcglobal.com> wrote in article
<35d038c...@news.skynet.be>...
> Hi all,
>
> I am working on a data warehousing project where we consider using
> sybase adaptive server IQ for our datamart solutions.
> I am looking for feedback of anyone who has used the product under NT
> or any other platform.
> Issues which I am interested in are (among others):
> - Scalability vs. platform issues (what about NT?)
> - Memory requirements
> - Performance issues
> and any other issues you can think of.
>
Steve Etchelecu
>
> It's also a good
> idea to partition your indexspaces in order to reduce I/O.
On Sun platforms, particularly ES6000, I would _highly_ recommend Sun's
A5000 Network Arrays (a.k.a., SENA and the photon). They are
phenomenal when it comes to I/O. I/O bound tasks go from hours to
minutes.
As an added bonus you're working with Volume Manager which is not as
good as IBM's AIX lvm but it's pretty good.
Steve Etchelecu
Diederik Dellaert
Thanks for the response,
Your response times seem on NT to be much better than the ones in our
"test"-environment. We are using a Compaq 5500R-two CPU (pentium pro
200) with one Gb of RAM. Queries on a 9 Million record fact table run
OK but as soon as we start to join these tables to our (evan small)
dimension tables, performance falls considerably.
Are you using more CPU or is it just the memory that makes the
difference.
Best Regards,
Diederik