Re: Blade vs Superdome

Let me agree with Andrew. Blades will displace large scale SMP systems for
many customers. Efficiently scaling a single system image and applications
to take advantage of very large CPU counts is hard to do - and we have more
than a few who have the need and the abilty to do so. But many large
systems are effectively (if not really) partitioned into smaller subsets of
CPUs.

From a VMS perspective, I would consider a blade system as a cluster in a
box with better interconnect technology and packaging. It's a pretty good
technology fit. It plays well into how VMS clusters are managed, and as a
high-reliabilty solution.



"Andrew" <andrew_harrison@xxxxxxxxxxxx> wrote in message
news:1140695218.529831.76130@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

JF Mezei wrote:
IBM just announced a new Blade cabinet.

##
IBM's new 15.75-inch-tall BladeCenter H chassis is 3.5 inches taller
than its predecessor. However, the design increases internal data
transfer capacity tenfold to 40 gigabits per second; adds faster
InfiniBand and Ethernet networking options; and includes more
sophisticated self-management features. The system still accommodates as
many as 14 blade servers, Big Blue said.
##

I assume that HP's offering will pretty much match this.


Do such blades start to approach the performance of the interconnects
between CPUs in Superdome systems ? Or are they still orders of
magnitudes slower (in terms of interconnects between CPUs only, not
actual CPUs).


They are still orders of magnitude slower than the interconnects used
on large SMP systems. All of the large SMP servers (HP, IBM and Sun)
have bisectional bandwidths of 50+ GB/s.


Or let me rephrase the question: Will Blade architectures
eventually/soon scale to compete/replace Superdome class systems ?

Blades are getting bigger, by that I mean that the blade building
blocks are getting larger. Sun for example is going to release an 8
Module 16 core AMD based blade.

Many of the big SMP systems are deployed using some form of hardware
partitioning and so you will find that the blades eat into this over
time. Blades give you access to some of the shared resources you get on
a large SMP system and if your maximum domain/LPAR etc partition does
not exceed the largest available blade in your chassis then you may
well see the blades replacing the larger SMP systems.



.

Relevant Pages

  • Re: Blade vs Superdome
    ... IBM's new 15.75-inch-tall BladeCenter H chassis is 3.5 inches taller ... Do such blades start to approach the performance of the interconnects ... between CPUs in Superdome systems? ... Many of the big SMP systems are deployed using some form of hardware ...
    (comp.os.vms)
  • Re: OT: Sunīs AMD Opteron
    ... > Fabio Cardoso wrote: ... > IA64 is ill suited for blades and large numbers of processors in a huge room. ... > And then, when you get into Galaxy class machines, you can also move CPUs ...
    (comp.os.vms)
  • HT apparently not detected properly on 2.4.23
    ... The same config with 2.4.23 only shows 2 cpus, ... properly detected on several other SMP systems I have at work (compaq ... send the line "unsubscribe linux-kernel" in ...
    (Linux-Kernel)
  • SMP & memory question
    ... Question regarding how SMP systems allocate memory... ... If there's a server with dual CPUs and 6GB of RAM total, ... Is it split 50/50 between the CPUs with each getting 3GB of RAM? ...
    (comp.os.linux.hardware)
  • Re: OpenVMS, Itanium and Blades...
    ... > Do you know if HP will develop Itanium Blades to run OpenVMS? ... Licensing would also have to be made such that 64 independant machines ... partitioned into galaxiy styles offering 4 nodes of 16 cpus each. ...
    (comp.os.vms)