Re: Massive Number of Concurrent Users
From: Niall Litchfield <niall.litchfield_at_gmail.com>
Date: Fri, 8 Jun 2012 18:20:08 +0100
Message-ID: <CABe10sb++zUzwzYd4aFm_a1o8Gq80v6M4t_zNQuJCmrOWoHKGQ_at_mail.gmail.com>
What sort of application is it, and what is the definition of concurrent users? You put an upper bound of 5000 on this figure. If this means 'concurrently connected' and this is real OLTP then I'd use 10% of that as an active workload or 500 active user processes. That doesn't equate to huge hardware demands, in fact and especially if workload can be segregated would probably make an excellent 3 node 4 proc RAC install on intel. If 5000 concurrent users means 5000 concurrently active sessions (say 50000 users for OLTP) then we're beyond mailing list advice IMO. Mixed mode OLTP/DSS apps (what I call Real Applications!) are also heavily dependent on actual usage patterns for analysis. You likely also want to read Neil Gunther's Universal Scalability Law as well IMO On Jun 8, 2012 5:56 PM, "Dave" <user4test_at_gmail.com> wrote:
> Just wondering if anyone on the list who has worked with a concurrent peak
> user load in the thousands (1000 - 5000) would mind sharing some nuggets on
> the hardware they used to support such a load. The load would come from an
> OLTP (80/20 r/w) web application using JBOSS application pools to the
> Oracle 11gR2 Ent database. I've been doing some research looking at
> powerful systems capable of supporting 80 Intel cores per system and
> connecting that with some PCIe Flash Memory for the L2 cache (flash cache)
> as well as using shared storage into a RamSan-630 via FC in order to
> maximize the use of the CPU cores and fast I/O. This would likely be a RAC
> scenario with two like servers. By the way, I am aware of how relatively
> "cheap" (cough) the hardware is compared to the millions in Oracle
> licensing this will cost. Still, I suspect the entire thing will be
> cheaper overall than an Exadata x2-8 system.
> Thanks!
>
>
> --
> http://www.freelists.org/webpage/oracle-l
>
>
>
Date: Fri, 8 Jun 2012 18:20:08 +0100
Message-ID: <CABe10sb++zUzwzYd4aFm_a1o8Gq80v6M4t_zNQuJCmrOWoHKGQ_at_mail.gmail.com>
What sort of application is it, and what is the definition of concurrent users? You put an upper bound of 5000 on this figure. If this means 'concurrently connected' and this is real OLTP then I'd use 10% of that as an active workload or 500 active user processes. That doesn't equate to huge hardware demands, in fact and especially if workload can be segregated would probably make an excellent 3 node 4 proc RAC install on intel. If 5000 concurrent users means 5000 concurrently active sessions (say 50000 users for OLTP) then we're beyond mailing list advice IMO. Mixed mode OLTP/DSS apps (what I call Real Applications!) are also heavily dependent on actual usage patterns for analysis. You likely also want to read Neil Gunther's Universal Scalability Law as well IMO On Jun 8, 2012 5:56 PM, "Dave" <user4test_at_gmail.com> wrote:
> Just wondering if anyone on the list who has worked with a concurrent peak
> user load in the thousands (1000 - 5000) would mind sharing some nuggets on
> the hardware they used to support such a load. The load would come from an
> OLTP (80/20 r/w) web application using JBOSS application pools to the
> Oracle 11gR2 Ent database. I've been doing some research looking at
> powerful systems capable of supporting 80 Intel cores per system and
> connecting that with some PCIe Flash Memory for the L2 cache (flash cache)
> as well as using shared storage into a RamSan-630 via FC in order to
> maximize the use of the CPU cores and fast I/O. This would likely be a RAC
> scenario with two like servers. By the way, I am aware of how relatively
> "cheap" (cough) the hardware is compared to the millions in Oracle
> licensing this will cost. Still, I suspect the entire thing will be
> cheaper overall than an Exadata x2-8 system.
> Thanks!
>
>
> --
> http://www.freelists.org/webpage/oracle-l
>
>
>
-- http://www.freelists.org/webpage/oracle-lReceived on Fri Jun 08 2012 - 12:20:08 CDT