How many physical processors we need to support 32K THREADS in IDS.

Hi Team,

Informix IDS supports upto 32K threads according to
most IDS realease notes.
My question is What sort of a machine you are looking
at to support this many
threads??
This came to my mind when considering max. num of
threads(or connections that matter) that each CPU or
NET VP process can support.
Because we can specify a number of connections as 999
through onmonitor
for each CPU or NET type virtual processor, this means
it exceeds max. value for
this parameter,256 occording to IDS release notes.
If we got a machine with 10 physical processors, the
Max CPU VPs we can allocate is 9.
That means maximum no of connections(considering each
connection supports only one thread) it can support is
= 256*9 =1994
To support 32K threads we need min 32000 *10/1994 =
160 CPU OR NET processes.
That means atleast 159 Physical processors.
Where do you find this sort of a machine ??
If I am wrong in calculating this value, what is the
correct way.
That leads us to the simple question,
How many physical processors we need to support 32K
threads??

Any thoughts.


Regards,
SPUTNIK



__________________________________
Do you Yahoo!?
Yahoo! Mail - Helps protect you from nasty viruses.
http://promotions.yahoo.com/new_mail
sending to informix-list

Julia M wrote:
> Informix IDS supports upto 32K threads according to
> most IDS realease notes.

So, your platform is Windows?

> My question is What sort of a machine you are looking
> at to support this many threads??

A big one. Why do you want to run that many threads? What are you
hoping to achieve? Do you want 32K concurrent connections to a single
DBMS instance?

It is more of a theoretical limitation -- if you have the h/w to run
it, then you reach a s/w limit at 32 K threads.

> This came to my mind when considering max. num of
> threads(or connections that matter) that each CPU or
> NET VP process can support.
> Because we can specify a number of connections as 999 through
> onmonitor for each CPU or NET type virtual processor, this means it
> exceeds max. value for this parameter, 256 occording to IDS release
> notes.

I think that you are complaining that ON-Monitor permits you to enter
a value up to 999 in the NETTYPE entries, whereas the release notes
indicate that the maximum is 256? Offhand, I don't know which, if
either, limit is correct - I haven't had a chance to configure that
big of a system recently. OTOH, you usually don't want to overload
any of the VPs. Note that CPU VPs process SQL; NET VPs process
connections.

> If we got a machine with 10 physical processors, the
> Max CPU VPs we can allocate is 9.

I think 'can' is too strong; you can allocate as many CPU VPs as you
wish, but we recommend that you don't allocate more than NUMCPUS - 1
(or fewer than 1).

> That means maximum no of connections (considering each
> connection supports only one thread) it can support is
> = 256*9 =1994
> To support 32K threads we need min 32000 *10/1994 =
> 160 CPU OR NET processes.

Well, following your nomenclature, the way I'd do that calculation is
that each VP can handle 256 threads, so 32K / 256 = 128. So, you need
a nice machine with 129 CPUs according to your understanding.

> That means at least 159 Physical processors.

161? Though 161 is at least 159 too :-)

> Where do you find this sort of a machine ??

Rather specialized labs, and not usually running Windows.

> If I am wrong in calculating this value, what is the
> correct way.
> That leads us to the simple question,
> How many physical processors we need to support 32K
> threads??

This comes back to my original question - well, second question.
Why do you need to worry about having 32K threads?

If you are going to be running 32K users concurrently against a single
DBMS, you definitely need a meaty machine. Big in all senses. I'm
not convinced said machine would be running Windows, either, but that
could just be bigotry. A 64-way box with say 1 TB main memory might
be sufficient - if you can configure enough connections on it. If
you're using network connections (and with 32K concurrent users, you
are using network connections), then if you say 'NETTYPE soctcp,2,99'
(is that complete - memory is rusty and MacOS X is giving me a hard
time right now and won't let me launch a PDF file; weird - I foresee a
reboot), then it will automatically allocate more listeners as needed.

--
Jonathan Leffler #include <disclaimer.h>
Email: jleffler@earthlink.net, jleffler@us.ibm.com
Guardian of DBD::Informix v2003.04 -- http://dbi.perl.org/

The recommendations for the network sizing is performance oriented for TCP
connections. For SHM connections, it is physical because the SHM connection
requires physical memory (IPCSHM). With one TCP network interface, you
could go well beyond the 256 limit. Also, if you have fragmentation and are
using PDQ, then each user thread will spawn additional threads. So it is
very possible, say, for a 100 user threads to spawn 1000 threads, once PDQ
is invoked. And it is also possible that 3200 user connections to spawn
32000 threads as well.

M.P.

"Julia M" <juliamrus@yahoo.com> wrote in message
news:ci6g42$ig5$1@news.xmission.com...
>
> Hi Team,
>
> Informix IDS supports upto 32K threads according to
> most IDS realease notes.
> My question is What sort of a machine you are looking
> at to support this many
> threads??
> This came to my mind when considering max. num of
> threads(or connections that matter) that each CPU or
> NET VP process can support.
> Because we can specify a number of connections as 999
> through onmonitor
> for each CPU or NET type virtual processor, this means
> it exceeds max. value for
> this parameter,256 occording to IDS release notes.
> If we got a machine with 10 physical processors, the
> Max CPU VPs we can allocate is 9.
> That means maximum no of connections(considering each
> connection supports only one thread) it can support is
> = 256*9 =1994
> To support 32K threads we need min 32000 *10/1994 =
> 160 CPU OR NET processes.
> That means atleast 159 Physical processors.
> Where do you find this sort of a machine ??
> If I am wrong in calculating this value, what is the
> correct way.
> That leads us to the simple question,
> How many physical processors we need to support 32K
> threads??
>
> Any thoughts.
>
>
> Regards,
> SPUTNIK
>
>
>
> __________________________________
> Do you Yahoo!?
> Yahoo! Mail - Helps protect you from nasty viruses.
> http://promotions.yahoo.com/new_mail
> sending to informix-list