Re: OT: high frequency trading

From: <Keegan.Holley_at_sungard.com>
Date: Sat, 13 Feb 2010 13:22:36 -0500

Oh the stories I could tell... It's times like this that I regret
posting under my real name. I'm sure the engineering dept in the average
investment bank can and has convinced management to waste money. I wasn't
necessarily trying to start an argument. I just haven't seen any data
about trading applications being able to move fast enough to make these
extreme low latency environments as useful as people say they are. All
the stats I've seen gauge value in terms of latency or power consumption
or throughput. I'm just curious if a complete comparison was done before
everyone started racing for extreme performance. The company I work for
actually develops such software so I'll look into the software aspect
further. Again not saying it's impossible, just that I have yet to see
data. I also have it on pretty good authority that at least one major
exchange does not care about their customers networks and all those low
latency orders will sometimes bottleneck if going to the same place at
the same time.

From:
Abdul <rslab007_at_gmail.com>
To:
Keegan.Holley_at_sungard.com
Cc:
Anthony Bonilla <anthonybonilla.ccie_at_gmail.com>, "ccielab_at_groupstudy.com"
<ccielab_at_groupstudy.com>, Nahskur Udniraht <expertinternetwork_at_gmail.com>,
Gregory Gombas <ggombas_at_gmail.com>, "Joseph L. Brunner"
<joe_at_affirmedsystems.com>, nobody_at_groupstudy.com
Date:
02/13/2010 01:09 PM
Subject:
Re: OT: high frequency trading

Keegan,
I won't get into a argument with you about this. Do you work for a
financial services company that does trading? Trading decisions made by
humans? Really? Have you been to the CME or NASDAQ recently? How many
human traders do you see in the pits vs say 5 or 10 years ago?

All I will say is that all that I mentioned is not technology for
technology sake. Financial services companies are making real money based
on trading strategy and low latency. Nobody at least not in the financial
world would spend money just for technology sake. Its spent if it gives
you an edge and bottom-line, helps you win in the market space. I can tell
you first hand, low latency technologies and designs have won, and won big
in the market place for us and quite frankly many other firms. This is not
a gimmick of "technology for technology sake".

But thats why I referenced my emails with, this only relates to Financial
services companies or High Computing environments. Most network services
are unconcerned with this level of latencies.

Anyway, I'm done talking about it. Thought I'd might add to a conversation
I know very well, and is the driver of articles like the one Network World
wrote in January or about 10Gb switches. Its the driver on why Cisco is
coming out with a lower latent chipset for the Nexus platform later on
this year.

This is real stuff happening in the network engineering world for whomever
is working or will work in Financial services or trading environments.

On Sat, Feb 13, 2010 at 12:37 PM, <Keegan.Holley_at_sungard.com> wrote:

I'm still not convinced that a trading application can serve up trades
that fast. Also, many of the trading decisions are made by humans and we
haven't even reached the "seconds" world. Infiniband is just another
gadget as is 10G and the inevitable 100G. A normal bus usually operates
at about 8-10G anyway. You're still thinking in terms of communication
between nodes. What about actually generating and analyzing data?
 Storing it? Backing it up? We are of course assuming that the actual
exchange can receive and process all this data at the speed at which it is
chucked at them without causing bottlenecks. We're also assuming that they
care. I'm not saying that it's impossible, but I would just like to see
a real world comparison done that shows a 2ms trading house vs. a 10ms
trading house in the same colo. Or even a trading house that saw real
world improvements (like those measured in dollars) after lowering their
latency to an exchange. My gut says that most of this is technology for
it's own sake.

From:
Abdul <rslab007_at_gmail.com>
To:
Keegan.Holley_at_sungard.com
Cc:
Nahskur Udniraht <expertinternetwork_at_gmail.com>, Anthony Bonilla <
anthonybonilla.ccie_at_gmail.com>, "ccielab_at_groupstudy.com" <
ccielab_at_groupstudy.com>, Gregory Gombas <ggombas_at_gmail.com>, "Joseph L.
Brunner" <joe_at_affirmedsystems.com>, nobody_at_groupstudy.com
Date:
02/13/2010 12:23 PM
Subject:
Re: OT: high frequency trading
Sent by:
<nobody_at_groupstudy.com>

Keegan,
You may a great point. And frankly your point is Cisco's argument why the
Nexus platform should still be considered for the High Frequency/Low
Latency
trading. So you right, .. once you've squeezed all the microseconds out of
the switch, and your "race to 0" is complete, it makes no sence to be
2microseconds better from one switch to another when the application still
adds 1millisecond of latency (more like a couple of milliseconds or even a
second).

But trading strategies out there are addressing that with
numerous optimization techniques on the software & even server
architecture
side. Finally this is where if you application is still that slow..
Infiniband? Hummm... Enough said.

Finally.. be advised.. most of this "advantage" on the latency side is
really played in the 10Gb arena. How many companies are serving up 10Gb of
data for consumption? Yeah, yeah, heard the serialization effect between
10Gb and 1Gb but seriously if your only pushing 1Gb of date on a 10Gb nic,
how much latency are you truly saving to the application?

Great topic, but really only Financial Services and High
computing environments are that concerned with shaving off microseconds.
For
the rest of the world, if you get your network to be milliseconds faster,
your the man. Cause their applications are still running in the seconds
world. :-)

On Sat, Feb 13, 2010 at 11:48 AM, <Keegan.Holley_at_sungard.com> wrote:

> Can you're software really move that fast? I understand the ability to
> move data faster and oversubscription rates. I simply have doubts that
> this translates to real world results.
>
>
>
>
>
> From:
> Nahskur Udniraht <expertinternetwork_at_gmail.com>
> To:
> Keegan.Holley_at_sungard.com
> Cc:
> "Joseph L. Brunner" <joe_at_affirmedsystems.com>, Anthony Bonilla
> <anthonybonilla.ccie_at_gmail.com>, "ccielab_at_groupstudy.com"
> <ccielab_at_groupstudy.com>, Gregory Gombas <ggombas_at_gmail.com>,
> nobody_at_groupstudy.com
> Date:
> 02/13/2010 08:29 AM
> Subject:
> Re: OT: high frequency trading
> Sent by:
> <nobody_at_groupstudy.com>
>
>
>
> network comes to play after you break 1 millisecond barrier ...
>
> we are using 4948 most of the time ...
>
> On Sat, Feb 13, 2010 at 6:50 PM, <Keegan.Holley_at_sungard.com> wrote:
>
> > This has been done and re-done over and over. You can colo at the
> > exchange itself or there are probably a wealth of carrier hotels
within
> > 10ms of it. The bottleneck is almost always going to be the software
> > though. I haven't actually seen studies on this, but off the top of
my
> > head I'm curious about the benefit of lowering latency from 15ms to
say
> 2
> > or 3. The software can take 1 or 2 full seconds or more to do it's DB
> > calls and actually use the connection.
> >
> >
> >
> >
> > From:
> > "Joseph L. Brunner" <joe_at_affirmedsystems.com>
> > To:
> > Gregory Gombas <ggombas_at_gmail.com>, Anthony Bonilla
> > <anthonybonilla.ccie_at_gmail.com>
> > Cc:
> > "ccielab_at_groupstudy.com" <ccielab_at_groupstudy.com>
> > Date:
> > 02/12/2010 10:54 AM
> > Subject:
> > RE: OT: high frequency trading
> > Sent by:
> > <nobody_at_groupstudy.com>
> >
> >
> >
> > Actually most shops send the orders to different exchanges and black
> books
> > and of course arbitrage between the price differences they can
exploit.
> > The can often find liquidity before anyone else knows it exists, and
> they
> > can send orders our for a very short time, of course pulling them if
> they
> > don't get the price they want...
> >
> > It's kind of a nice study to work with these guys- they do eat the
> slower
> > players lunch (that may be software not just location based slowness).
> >
> > Pretty much all the major players are already at the exchanges and
> > therefore you have to do it.
> >
> > -Joe
> >
> > -----Original Message-----
> > From: nobody_at_groupstudy.com [mailto:nobody_at_groupstudy.com] On Behalf
Of
> > Gregory Gombas
> > Sent: Friday, February 12, 2010 9:17 AM
> > To: Anthony Bonilla
> > Cc: ccielab_at_groupstudy.com
> > Subject: Re: OT: high frequency trading
> >
> > Tell them it won't matter anyway because whatever slight edge they
> > will get over their competitor by collocating at the exchange will
> > disappear once their competitor does the same :-)
> >
> > On Thu, Feb 11, 2010 at 10:23 PM, Anthony Bonilla
> > <anthonybonilla.ccie_at_gmail.com> wrote:
> > > Hi all, I am back again. Have a question regarding high frequency
> > trading.
> > > We are planning on collocating at an exchange for trading and are
> > looking
> > > for doing lowest latency possible. I wanted to see if anyone else
is
> > doing
> > > this and if there are any recommendations. I am currently thinking
> > about
> > > 4900M and nexus 5k (layer 2) but am interested in seeing what others
> > have
> > > done and whether there are any best practices from cisco to ensure
> that
> > we
> > > achive lowest latency. TIA.
> > >
> > >
> > > Blogs and organic groups at http://www.ccie.net
> > >
> > >
> _______________________________________________________________________
> > > Subscription information may be found at:
> > > http://www.groupstudy.com/list/CCIELab.html
> >
> >
> > Blogs and organic groups at http://www.ccie.net
> >
> >
Received on Sat Feb 13 2010 - 13:22:36 ART

This archive was generated by hypermail 2.2.0 : Mon Mar 01 2010 - 06:28:35 ART