Discussion:
The importance of accurate time keeping and latency
Add Reply
Dirk Munk
2017-04-03 11:56:57 UTC
Reply
Permalink
Raw Message
We have had several discussion about accurate timekeeping and latency in
this group lately.

Let me add a nice excerpt from a news article on this matter.

The BBC has a news article on its web site, called "Tick tock: The
importance of knowing the right time". It is written by Tim Hadford, who
is also writing for The Financial Times.

The excerpt reads:

"Some financiers recently calculated it was worth spending $300 million
drilling through mountains between Chicago and New York to lay
fibre-optic cables in a slightly straighter line. That sped up
communication between the two cities' exchanges by three milliseconds."
Kerry Main
2017-04-03 12:57:05 UTC
Reply
Permalink
Raw Message
-----Original Message-----
Dirk Munk via Info-vax
Sent: April 3, 2017 7:57 AM
Subject: [Info-vax] The importance of accurate time keeping and
latency
We have had several discussion about accurate timekeeping and
latency in this group lately.
Let me add a nice excerpt from a news article on this matter.
The BBC has a news article on its web site, called "Tick tock: The
importance of knowing the right time". It is written by Tim Hadford,
who is also writing for The Financial Times.
"Some financiers recently calculated it was worth spending $300 million
drilling through mountains between Chicago and New York to lay
fibre-
optic cables in a slightly straighter line. That sped up
communication
between the two cities' exchanges by three milliseconds."
Volume matters .. billions (trillions?) of transactions over X time x
3msec is a big deal.

Especially when NY is currently the centre of stock exchange trades.

On a semi-related "importance of time" note, there is an update in the
latest VSI roadmap for support of Precision Time Protocol (P2P) -
V8.x, Q1 2018.


Regards,

Kerry Main
Kerry dot main at starkgaming dot com
j***@yahoo.co.uk
2017-04-03 14:45:22 UTC
Reply
Permalink
Raw Message
Post by Dirk Munk
We have had several discussion about accurate timekeeping and latency in
this group lately.
Let me add a nice excerpt from a news article on this matter.
The BBC has a news article on its web site, called "Tick tock: The
importance of knowing the right time". It is written by Tim Hadford, who
is also writing for The Financial Times.
"Some financiers recently calculated it was worth spending $300 million
drilling through mountains between Chicago and New York to lay
fibre-optic cables in a slightly straighter line. That sped up
communication between the two cities' exchanges by three milliseconds."
Related article:
http://www.bbc.co.uk/news/business-39129620

Tim Harford (nb spelling) is the Undercover Economist at the
Financial Times and also presents the generally excellent
BBC Radio 4 programme "More or Less", which is "the BBC's
guide to statistics in the news and in life" - a co-production
with the Open University. Lots of "more or less" related stuff
on the web, much of which should be available outside the UK.

I suspect that in the case of financial trading etc it's being
3 milliseconds (or whatever) *faster than the competition* that
matters, not specifically the absolute timing. When something
hits the market wires, and you're 5ms behind your competition
in responding (buying/selling before the price changes much),
you're nowhere, at least in the world of high performance
trading.

If you can be 3ms ahead of your competition by spending huge
amounts of (other people's) money that the competition can't
afford, e.g. to build a 'shorter' network route that others
can only access for an extortionate fee, well what's not to
like?

Other less drastic options are available but may not be
directly relevant to the normal business of comp.os.vms...
Bob Gezelter
2017-04-03 16:34:27 UTC
Reply
Permalink
Raw Message
Post by j***@yahoo.co.uk
Post by Dirk Munk
We have had several discussion about accurate timekeeping and latency in
this group lately.
Let me add a nice excerpt from a news article on this matter.
The BBC has a news article on its web site, called "Tick tock: The
importance of knowing the right time". It is written by Tim Hadford, who
is also writing for The Financial Times.
"Some financiers recently calculated it was worth spending $300 million
drilling through mountains between Chicago and New York to lay
fibre-optic cables in a slightly straighter line. That sped up
communication between the two cities' exchanges by three milliseconds."
http://www.bbc.co.uk/news/business-39129620
Tim Harford (nb spelling) is the Undercover Economist at the
Financial Times and also presents the generally excellent
BBC Radio 4 programme "More or Less", which is "the BBC's
guide to statistics in the news and in life" - a co-production
with the Open University. Lots of "more or less" related stuff
on the web, much of which should be available outside the UK.
I suspect that in the case of financial trading etc it's being
3 milliseconds (or whatever) *faster than the competition* that
matters, not specifically the absolute timing. When something
hits the market wires, and you're 5ms behind your competition
in responding (buying/selling before the price changes much),
you're nowhere, at least in the world of high performance
trading.
If you can be 3ms ahead of your competition by spending huge
amounts of (other people's) money that the competition can't
afford, e.g. to build a 'shorter' network route that others
can only access for an extortionate fee, well what's not to
like?
Other less drastic options are available but may not be
directly relevant to the normal business of comp.os.vms...
John,

A good point. Disparate latencies are a problem in markets.

A nicely readable article on the problem appeared in The New York Times magazine a while back. See https://www.nytimes.com/2014/04/06/magazine/flash-boys-michael-lewis.html

Hint: As was noted in "The Sum of All Fears" the way to get a pulse to arrive at the same time through disparate wires is to cut the wires to the same length.

- Bob Gezelter, http://www.rlgsc.com
j***@yahoo.co.uk
2017-04-03 16:36:41 UTC
Reply
Permalink
Raw Message
Post by j***@yahoo.co.uk
Post by Dirk Munk
We have had several discussion about accurate timekeeping and latency in
this group lately.
Let me add a nice excerpt from a news article on this matter.
The BBC has a news article on its web site, called "Tick tock: The
importance of knowing the right time". It is written by Tim Hadford, who
is also writing for The Financial Times.
"Some financiers recently calculated it was worth spending $300 million
drilling through mountains between Chicago and New York to lay
fibre-optic cables in a slightly straighter line. That sped up
communication between the two cities' exchanges by three milliseconds."
http://www.bbc.co.uk/news/business-39129620
Tim Harford (nb spelling) is the Undercover Economist at the
Financial Times and also presents the generally excellent
BBC Radio 4 programme "More or Less", which is "the BBC's
guide to statistics in the news and in life" - a co-production
with the Open University. Lots of "more or less" related stuff
on the web, much of which should be available outside the UK.
I suspect that in the case of financial trading etc it's being
3 milliseconds (or whatever) *faster than the competition* that
matters, not specifically the absolute timing. When something
hits the market wires, and you're 5ms behind your competition
in responding (buying/selling before the price changes much),
you're nowhere, at least in the world of high performance
trading.
If you can be 3ms ahead of your competition by spending huge
amounts of (other people's) money that the competition can't
afford, e.g. to build a 'shorter' network route that others
can only access for an extortionate fee, well what's not to
like?
Other less drastic options are available but may not be
directly relevant to the normal business of comp.os.vms...
Grrr. Forgot a piece... memory error correction required.

Tim probably simplified his story for the benefit of the
Radio 4 audience. And why not. But that's not us.

Fast trading is one of a number of classes of application,
some niche and some not so niche, some low value and some
not so low value, that no sensible organisation will want
to put in any kind of "cloud". One size does not fit all.

*Now* hopefully returning you to normal comp.os.vms.
IanD
2017-05-02 15:28:03 UTC
Reply
Permalink
Raw Message
High frequency trading does indeed rely upon reducing latency in order to get an advantage. Isn't this why the likes of Goldman Sachs set to ship as close as physically possible to the NYSE so as to gain an advantage in reducing latency

So proximity to the data is what's important as physical distance directly correlates to latency

What happens if say the NYSE moves to the cloud? I don't know if at some point in the future the likes of AWS gets good enough so as to offer a guaranteed compute response time. You can already but a dedicated non shared instance if you want one

To say that there are industries that will never move to the cloud (I'm not saying you did say this) is to try and project today's compute environment forever forward, which would be folly

If superlarge memory machines like The Machine ever eventuate, then you may indeed see clustering of companies happen around technology. Pays ya money to get a compute response time advantage etc. Suddenly proximity becomes a tradable commodity

My ACM magazine turned up and there is an article on time accuracy in there and the need for increasing accuracy to cope with transactions and operations occurring faster and faster as latency is being shortened. I'm hoping to get a chance to read it and I'm also hoping that the new VMS contains enough flexibility within its core to be able to handle such things as finer time granularities etc

It really does seem that the OS of today needs to be able to have every aspect of itself configurable to handle the needs of tomorrow
IanD
2017-05-02 15:32:00 UTC
Reply
Permalink
Raw Message
High frequency trading does indeed rely upon reducing latency in order to get an advantage. Isn't this why the likes of Goldman Sachs set up shop as close as physically possible to the NYSE so as to gain an advantage in reducing latency

So proximity to the data is what's important as physical distance directly correlates to latency

What happens if say the NYSE moves to the cloud? I don't know if at some point in the future the likes of AWS gets good enough so as to offer a guaranteed compute response time. You can already but a dedicated non shared instance if you want one

To say that there are industries that will never move to the cloud (I'm not saying you did say this) is to try and project today's compute environment forever forward, which would be folly

If superlarge memory machines like The Machine ever eventuate, then you may indeed see clustering of companies happen around technology. Pays ya money to get a compute response time advantage etc. Suddenly proximity to a data source becomes a tradable commodity

My ACM magazine turned up and there is an article on time accuracy in there and the need for increasing accuracy to cope with transactions and operations occurring faster and faster as latency is being shortened. I'm hoping to get a chance to read it and I'm also hoping that the new VMS contains enough flexibility within its core to be able to handle such things as finer time granularities etc

It really does seem that the OS of today needs to be able to have every aspect of itself configurable to handle the needs of tomorrow
Kerry Main
2017-05-06 02:21:48 UTC
Reply
Permalink
Raw Message
-----Original Message-----
via Info-vax
Sent: May 2, 2017 11:32 AM
Subject: Re: [Info-vax] The importance of accurate time keeping and
latency
High frequency trading does indeed rely upon reducing latency in order
to get an advantage. Isn't this why the likes of Goldman Sachs set up
shop as close as physically possible to the NYSE so as to gain an
advantage in reducing latency
So proximity to the data is what's important as physical distance directly
correlates to latency
What happens if say the NYSE moves to the cloud? I don't know if at
some point in the future the likes of AWS gets good enough so as to
offer a guaranteed compute response time. You can already but a
dedicated non shared instance if you want one
Do not get caught up in the industry public cloud washing hype.

Public Cloud is simply another term for "outsourcing-lite". The capacity on demand is simply a gui added on to a service catalogue with some workflow automation behind the scenes.

There are commercial add-on products you can buy today to layer on top of VMware to get very similar results by building an internal private cloud.

[snip]

So, put another way, your question really reads " What happens if say the NYSE moves to outsource their IT to an outsourcing vendor who has SysAdmins from around the world maintaining their IT?"

Since NYSE and other mission critical environments like this are heavily, heavily customized with highly integrated proactive call and escalation flows, the chances of NYSE outsourcing their infrastructure to another vendor like AWS or Azure who have little experience in their business is slim to none. In addition, latency would go through the roof because while Cloud vendors will typically provide SLA's promising high availability (albeit with few, if any penalties), they typically do not provide latency guarantees.

Heck, AWS, Azure would likely not even want that business given the extra costs they would have to incur.

😊


Regards,

Kerry Main
Kerry dot main at starkgaming dot com

Loading...