Small sample of ARCA FAST data
- Thread starter nitro
- Start date
Call the exchange. They usually have a few samples lying around.
In terms of development time, with ARCA, it's probably quicker and more practical to just roll out the uncompressed version, though I don't know your organization's requirements.
Quote from nitro:
Why didn't I think of that? Thanks
Why program to old technology? If it is worth doing, it is worth doing right, imo.
The FAST protocol takes longer to decode since you're serially iterating over what's going on with the template-specified presence map and looping until you hit the stop bit.
If you have very high throughout and low latency, I don't think you get huge benefits from FAST, although I wouldn't mind seeing numbers if you have them. Care to share?
If you want quick classifiers or a quick filter for dropping products you don't care about, you end up incurring this decode latency for bullshit products, right?
I'd like to mention that I could be completely wrong. Depends on how much your effective throughput is, I guess. If FAST compacts lots of packets quickly and gets it down the wire faster, there could be less latency on account of the switches and smaller packet sizes.
Please share numbers. ;-)
It has been quite a while since our system went live on Arca, so I hope this still works ...
Edit: Sorry, I don't know if I am allowed to post this Sample data site. I will PM nitro directly.
It contains a capture of ARCA multicast book data. Which I believe is what you are looking for.
ARCA's messaging format is not FAST. ARCA is FAST "like", with custom value based tags, etc.
As for performance, obviously i can not speak for other implementations, but our own custom implementation is close to Pantor's benchmark, which is around 10M msgs / decoded per sec. (no custom hardware acceleration). My own bench mark, on my 3Ghz thinkpad T60, is around 7.2M msgs / sec, full decode, and into internal data structure, ready to publish to the ticker plant. And this is much faster than our own parser for the flat format, which is only around 500 - 600k msg / sec. Our implementation is in C (not even C++, as those who have read my other posts will know).
Hope this helps.
Edit: Sorry, I don't know if I am allowed to post this Sample data site. I will PM nitro directly.
It contains a capture of ARCA multicast book data. Which I believe is what you are looking for.
ARCA's messaging format is not FAST. ARCA is FAST "like", with custom value based tags, etc.
As for performance, obviously i can not speak for other implementations, but our own custom implementation is close to Pantor's benchmark, which is around 10M msgs / decoded per sec. (no custom hardware acceleration). My own bench mark, on my 3Ghz thinkpad T60, is around 7.2M msgs / sec, full decode, and into internal data structure, ready to publish to the ticker plant. And this is much faster than our own parser for the flat format, which is only around 500 - 600k msg / sec. Our implementation is in C (not even C++, as those who have read my other posts will know).
Hope this helps.
They have a FAST channel where multicast data in FAST format gets spewed. Might depend on your subscription.
rufus_4000
Thanks again.
I got your PM. Thank you very much. I will run the decoder and see if it works.
Right, I think ARCA anticipated the FAST protocol and as a result they are not 100% compliant.
Those numbers are really impressive for a non harware-accelerated solution. In fact, they are the fastest numbers I have seen so far!
Thanks again.
Looks like Pantor reached 66M msgs / sec, they can process 8 GB/s out of 10 GB/s network interface. Rolf Anderssen (Pantor's CEO, former CTO of OMX and Stockholm Stock Exchange), also chairs the FIXFAST working group.
http://ipip.intel.com/go/wp-content/uploads/2009/03/iif_summer08.pdf
As far as we know, our approach is quite diff from Pantor's. But then again, we are not in the packaged software vendor business, heh.