Re: SQL active - active cluster hardware design?

I think what Linchi is trying to point out is that it is difficult to make
blanket statements about performance because performance depends to a very
large degree on your application. TPCC benefits from having the whole
database in buffer memory so 64 bit addressing provides a significant
performance boost because the large address space eliminates the AWE
overhead of 32 bit addressing. The larger procedure cache also helps TPCC
because it ensures that plans will always be in cache. On the other hand,
if the working set for your instance fits into 2 or 3 GB of buffer space and
you don't have enough plans to exhaust the procedure cache then 32 bit MAY
run faster than 64 bit because current 32 bit processor usually have high
clock speeds than 64 bit processors. We see this regularly when a customer
runs their own "benchmark" by executing the same statement a few thousand
time on a few hundred rows in an sample database and wants to know why their
new 64 bit processor is not as fast as a 32 bit one. My answer is usually
that if their application on has one SQL statement and a couple MB of data
them they're wasting their money on a 64 bit processor but if not, they
should try the test with their real application and data.

Working set for a database is very application dependent. Many applications
only work with a fairly limited part of the data regularly so the working
set may be 20 or 30 percent of the database size. On the other hand some
application regularly do queries that span the whole database so the
workingset may be close to 100% of the database size. If you have 8 5GB
databases, chances are pretty good that the workingset is significantly more
than 2 GB so a 64 bit processor with maybe 32GB or memory would likely hold
most of the workingset and give the best performance in this case.

--
This posting is provided "AS IS" with no warranties, and confers no rights.
Use of included script samples are subject to the terms specified at
http://www.microsoft.com/info/cpyright.htm

"Linchi Shea" <LinchiShea@xxxxxxxxxxxxxxxxxxxxxxxxx> wrote in message
news:84358BAF-29B6-458E-B465-ED31BD7B148A@xxxxxxxxxxxxxxxx
As far as 32 versus 64 bit computing, and IA64 versus x64, show me any
documented studies that proves your claim.

I don't know whether you can find any public documented studies as such.
But
I have done many tests myself and found that to be the case. Basically, I
have seen the results all over the map with different workloads (e.g.
cpu-bound, memory-bound, I/O-bound, etc) when I systematically compared
different platforms. I simply stated the facts as I saw them. Unfornately,
I
can't disclose all the test details in a public forum like this.

Here's mine: take a look at any of the TPM-C benchmarks and rank the
systems by both metrics and tell me which ones outperform.

Note that I didn't say 32-bit is always better than 64-bit or x64 is
always
better than Itanium. My replies were 'not necessarily' in both cases. So
even
if you can find TPC-C benchmarks to be better on Itanium, that doesn't
contradict what I said.

What specifically did you have in mind? What kind of workload did you
think
would perform better on 32-bit or x64 over IA64?

For x64 vs IA-64, as I said it depends a lot on how the system vendors
make
their systems with these processors (e.g. how they design the system
architecture including the interconnect among processors and memory
modules).
At the highest end, you may find IA-64 machines to be more powerful. But
for
the majority of business applications, x64 servers may be sufficient. And
if
we are talking about comparing servers at the same pricing level, I have
recently done the same tests on a cheap DL585 and a much more expensive
Itanium server with better memory configuration, and for these particular
test workloads (which inlcude a lot of BLOB manipulations), the 8-way
commodity DL585 beat that 8-way non-commodity Itanium box hands down.

Again, I stated only what I have had experience with. I did not, and would
not want to, say that x64 is always better than IA-64. It would be too
dangerous to make generalized statement like that.

Linchi

"Anthony Thomas" wrote:

I think you have just proved my point: no one could answer this
particular
installation question without a better understanding of the usage
requirements.

As far as 32 versus 64 bit computing, and IA64 versus x64, show me any
documented studies that proves your claim. Here's mine: take a look at
any
of the TPM-C benchmarks and rank the systems by both metrics and tell me
which ones outperform.

What specifically did you have in mind? What kind of workload did you
think
would perform better on 32-bit or x64 over IA64?


Anthony Thomas


--

"Linchi Shea" <LinchiShea@xxxxxxxxxxxxxxxxxxxxxxxxx> wrote in message
news:081C88DD-8A9E-485D-9746-5F93BB565597@xxxxxxxxxxxxxxxx
Is 64-bit better than 32-bit? Yes.
Not necessarily. It depends on the workload.

Is Itanium better than x64? Yes.
Not necessarily. It highly depends on the system design. We have found
cases
where an x64 box easily beats an Itanium box on the same workload.

Linchi

"Anthony Thomas" wrote:

You will need to know how you plan on using it: OLTP, DSS, Data
Warehouse,
Data Marts, OLAP?

You will need to know number of Transactions/Batch Requests per
second.

You will need to know number of users and databases. What are the
database
sizes?

Is more memory better? Yes.

Is more, larger processors? Yes.

Is 64-bit better than 32-bit? Yes. Is Itanium better than x64?
Yes.

However, what you need to build is what you need to process plus some
room
for future growth over the next 2 to 5 years; otherwise, you may end
up
under-powering, or over-paying for the system.

Sincerely,


Anthony Thomas


--

"gili98" <gili98@xxxxxxxxxxxxxxxxxxxxxxxxx> wrote in message
news:5ACB71E4-EFDF-4C03-BE17-E2D96964A467@xxxxxxxxxxxxxxxx
I want to make a two node active-active ms cluster. how to design
die
two
server hardware (RAM, processor). we use IBM xseries server.
is it good to use 64bit?

mfg








.

Relevant Pages

  • Re: SQL active - active cluster hardware design?
    ... I don't know whether you can find any public documented studies as such. ... have seen the results all over the map with different workloads (e.g. ... if you can find TPC-C benchmarks to be better on Itanium, ... For x64 vs IA-64, as I said it depends a lot on how the system vendors make ...
    (microsoft.public.sqlserver.clustering)
  • Re: Oracle statement of direction
    ... indifference to all things Clusters could not drive you to force loyal ... database to exploit fully the performance features of the new itanium ... While we at Rdb engineering will continue to ... both database and applications will run entirely ...
    (comp.os.vms)
  • RE: Downgrade from SQL2005 x64 EE to SQL2005 x64 SE
    ... We installed a new version of x64 SE on a new x64 bit platform and moved the ... "Mohit K. Gupta" wrote: ... MCITP: Database Administrator ... Was wondering if anyone could tell me the potential issues with downgrading ...
    (microsoft.public.sqlserver.setup)
  • Re: Database toolkit connection problems
    ... I'm using Vista Home Premium (x64) as the client to connect to a SQL Server 2008 instance which is on the same machine. ... restart Matlab and try to connect to the database in connection through the usual "database" command...this should fix the problem...it did for me. ... Good luck. ...
    (comp.soft-sys.matlab)
  • Re: DataBase For Download
    ... |> to where I can find sample database, which I could use for testing. ... I would generally call those workloads, ... send the line "unsubscribe linux-kernel" in ... More majordomo info at http://vger.kernel.org/majordomo-info.html ...
    (Linux-Kernel)