[Info-vax] OT: London stock exchange switches to Linux

[John Wallace]

On Jan 25, 11:27 am, Neil Rieck <n.ri... at sympatico.ca> wrote:
> London stock exchange switches to Linux
>
> http://www.theinquirer.net/inquirer/news/1588339/london-stock-exchang...
>
> Remember a couple of years back when the LSE tried to run their
> operation on Window Server 2003?
>
> NSR

There is a passing mention at the end of that article of the latest
fad in computerised trading, "high frequency trading". If you want to
read about what it is in market terms, there's plenty to read
elsewhere (though it may not be pleasant reading for those who think
the financial sector should serve its customers, rather than the other
way round).

There are also "reader comments" at the end of the Inquirer article.
So far, the commenters seem to think that VMS expertise is no longer
available. Can that be true, or is it just that the remaining VMS
folks have sufficiently strict criteria that City-oriented jobs aren't
appealing?

Anyway, in technical terms, high frequency trading needs a computing
platform with consistent low-millisecond-level latencies and response
times. Sometimes the compute workload is allegedly so intense that
part of it has to be offloaded to dedicated compute engines such as
FPGA cards and such. Obviously that kind of predictable responsiveness
makes Windows entirely inappropriate, any fool could have seen that
coming, but Billco managed to lunch enough Accenture and LSE staff to
fool them initially.

Windows may not have been up to the job, but a "soft real time" OS
such as a Linux with a pre-emptible kernel is probably entirely
capable of "good enough" performance. VMS could do it too, but that's
another story, one seemingly of insufficient interest to HP and its
"partners".

There's nothing particularly new in this LSE story, some folks have
been doing this kind of thing for a year or three. Like HP, for
example, and Novell/SuSe, in this 2008 press release (there were other
similar ones from other x86 folk before and after this one). The HP
group in this picture is of course the Proliant folks, with x86-64
blades. Note that this press release isn't a trading application, it's
a Reuters messaging thing, but there are similarities in the
requirements:

http://www.novell.com/news/press/novell-sets-new-performance-records-with-suse-linux-enterprise-real-time

Another SuSe example, this time from June 2009 on IBM x86-64 hardware,
talks about saving 30 microseconds latency in a UDP multicast
messaging application:
http://blog.29west.com/2009/06/23/lbm-latency-linux-10-gigabit-ethernet-udp-multicast-kernel-bypass-cisco-solarflare/

"The report tested throughput as well as latency. Using small messages
(64-bytes), 3 publishing applications on a single server could
generate almost 2.5 million messages per second. With larger (1204-
byte) messages, 3 publishing applications on a single server could
generate 550,000 messages per second driving the network to over 4.5
gbps."

Nice.

Based on what's gone before, I'd say the technology may well exist for
the LSE to do their migration within a year (JF's question). Whether
sufficient skilled people are available is a slightly different
question, but there seem to be a handful of companies claiming
competence in this space.

Next question: why would anyone want an x86 mainframe, even for a job
like this? Unisys did do them (their ES7000 line), and for a while
Compaq rebadged them as Proliant 9000. It's a tiny tiny market. Blades
and even rack mount servers have got what most enterprise customers
want, with the exception (for now) of an answer for those folks who
for some reason need single system image massive-memory massive-SMP
systems. Massive in this context currently means a workload that needs
more than today's biggest Proliant offering: iirc, 8 sockets (with 6
cores per socket ie 48CPUs) and 512GB of memory. For anything bigger
than that kind of system, HP's current offer is IA64 (for now).

Interesting times.