Today we announced the launch of a new MCFX quote server

Today we announced the launch of a new MCFX quote server in beta-test mode:
http://forum.tssupport.com/viewtopic.php?p=16261

Congratulations.

I just subscribed to MCFX Pro.

And I had cancel my eSignal Premier (+Forex) susbcription immediatly after seeing how fast the new MCFX backfill is ...

It's not a joke I realy did it ... eSignal is too expensive because of the GTIS Forex Feed including 200 contributors, forwars, money market, etc ... + Futures + Equities ... there is no cost effective "Spot Forex Only" product for the moment, so it's "eSignal Premier" or Nothing ...

So now this is my MCFX vs eSignal + Multicharts backfill Speed comparison on EURUSD and GBPUSD.

eSignal symbols : EUR@GFT A0-FX and GBP@GFT A0-FX (Please note that GFT datas backfill from GTIS is 20 times faster than the same coming from FXCM or composites on GTIS ... I don't know why ... so I launch a comparison with the best eSignal perfomances conditions )

Because only 6 months of data are available on eSignal I did 2 tests :

A. 6 Month backfill (pure data server Backfill on both)

B. 5 years backfill (with for Multicharts : 6 month of data server backfill + 4 1/2 years from ASCII imported historical datas from TS)

Results

6 Months of 1 Hour datas

- eSignal : 10 sec
- MCFX : 2 sec (5x faster)


5 Years of 1 Hour datas

- eSignal : 1min30 sec
- MCFX : 7 sec (12x faster ... but it's for fun because everybody know that Multichart local data performance is catastrophic)



Ok these difference are probably due to the Bar Compression process on the new MCFX server : if you ask for 1 hour bars, you receive 1 hour bars. On eSignal, you only receive 1 minutes localy compressed to generate 1 hour bars.

So this is a second test on 6 months data for different time frames :

6 Months of 1 minute datas

- eSignal : 10 sec (the same, see above)
- MCFX : 18 sec (1,8x slower)

6 Months of 5 minute datas

- eSignal : 10 sec (the same, see above)
- MCFX : 10 sec (exactly the same as eSignal)

6 Months of 15 minute datas

- eSignal : 10 sec (the same, see above)
- MCFX : 5 sec (2x faster)

6 Months of 30 minute datas

- eSignal : 10 sec (the same, see above)
- MCFX : 4 sec

So MCFX have exactly the same performance as eSignal for 5 min data. It is several times faster when the time scale increase.

I never use charts with time frames lower than 5 minutes, and I generaly use Daily, 1 Hour and 15 Min. New MCFX Server is a very big benefit for me and for major part of the traders.

So congratulations to the Developement team, I hope this Bar Compression algorithm will be included in the next Multichart for local data backfill

Hello Fabrice,

At this particular stage, the priority for MCFX data server is its reliability. We are planning to work on the speed of data loading by using better data compression methods - which will make data loading much faster.

As for MultiCharts, we do know that data loading is not very fast in the current version, but we wouldn't call it catastrophic. If you have come across a situation when data loading was unacceptably slow, could you please give us more details on what kind of data was loaded, over what period, etc. Also, we would still like to know what would be the practical implications of the immediate response when requesting large quantities of data.

And as I mentioned before, the loading in the upcoming MC version has been considerably improved.

Regards.

Thanks for keeping your promise to release this beta before the end of the year.

The improvement in loading speed is substantial, and very, very welcome.

The gappy data problems in the n-day charts seem to have been cured.

I haven't detected any RSI discrepancies with the new version.

There does seem to be missing data for December 14th in the hourly and daily charts for a number of currency pairs, which affects the medium-term indicators.

There was some unusual behaviour on Sunday night between market open (11pm local time) and midnight, but I haven't seen any recurrence.

At times of high volatility, there seems to be a considerable time lag (1 to 2 minutes) between the price displayed on MCFX and the price quoted at FXCM

M Klein

Dear M Klein,
Thanks for the pointing the gap problem. We will fix it in a few days.
You said:
"At times of high volatility, there seems to be a considerable time lag (1 to 2 minutes) between the price displayed on MCFX and the price quoted at FXCM "

Could you let me know your comparative approach?

We are aware of price discrepancies. Usually MCFX and FXCM Trading workstation quotes are off by 5 points. But this value is static and doesn’t change in volatile market. We found that if we restart FXCM Trading workstation several times we get the identical quotes. It looks like they connect to different servers and get different data. We are still investigating this issue and consulting with FXCM. If you have other problem please specify it.

Could you let me know your comparative approach?

I'm comparing spot price on the FXCM Trading Station package and the FXCM-feed charts at dailyfx.com with a 1-minute MCFX chart. The MCFX chart can lag by several minutes at certain times of day.

The MCFX timestamp for the most recent 1m bar sometimes shows a time several minutes behind the true time as displayed on the server clock. For example, today at 1630GMT there was a lag of about six minutes, diminishing to about 2 minutes by 1700GMT. At 1704GMT the lag has dropped to zero.

It's as if there's a queue or backlog that builds up at times of high volatility, and that backlog takes time to clear.

M Klein

Just one more thing for the moment.

The prior version of MCFX treated the week as lasting 5 days (Sunday open to Friday close), with any price movement on Sunday evening contributing to Monday's bar. The new version seems to allow for a 7 day week, with the result that price movement before 2200GMT on Sunday evening creates a new bar, containing little data. This affects the daily charts in particular. All daily charts now show "doji" bars for Sunday 30/12/2007 and Sunday 06/01/2008

N-day charts also now seem to include weekends as two trading days, so a 2-day chart has a doji bar for 05/01-06/01. This distorts the mid-term picture. It makes much more sense to treat the week as being composed of five days, as before.

M Klein

Could you let me know your comparative approach?

I'm comparing spot price on the FXCM Trading Station package and the FXCM-feed charts at dailyfx.com with a 1-minute MCFX chart. The MCFX chart can lag by several minutes at certain times of day.

The MCFX timestamp for the most recent 1m bar sometimes shows a time several minutes behind the true time as displayed on the server clock. For example, today at 1630GMT there was a lag of about six minutes, diminishing to about 2 minutes by 1700GMT. At 1704GMT the lag has dropped to zero.

It's as if there's a queue or backlog that builds up at times of high volatility, and that backlog takes time to clear.

M Klein

Thanks for the comments. Could you make a picture of the lag?

Just one more thing for the moment.

The prior version of MCFX treated the week as lasting 5 days (Sunday open to Friday close), with any price movement on Sunday evening contributing to Monday's bar. The new version seems to allow for a 7 day week, with the result that price movement before 2200GMT on Sunday evening creates a new bar, containing little data. This affects the daily charts in particular. All daily charts now show "doji" bars for Sunday 30/12/2007 and Sunday 06/01/2008

N-day charts also now seem to include weekends as two trading days, so a 2-day chart has a doji bar for 05/01-06/01. This distorts the mid-term picture. It makes much more sense to treat the week as being composed of five days, as before.

M Klein

We will see the problem and let you know. Thank you.

Thanks for the comments. Could you make a picture of the lag?

I'll try, but because the lag happens unpredictably and when the market is volatile, I tend to be very busy working, and there isn't always time to arrange the necessary windows to make a pretty picture. Unfortunately I'm not being paid to beta-test MCFX.

The other problem is that MCFX seems to corrupt the contents of the Windows clipboard, so that often I'm unable to produce a screen grab at all.

Today I observed a lag of around 4 minutes at about 1500GMT. I'll try to make a screen capture tomorrow.

M Klein

There was some unusual behaviour on Sunday night between market open (11pm local time) and midnight, but I haven't seen any recurrence.

This evening I've seen the same effect. Between NY close (2300 local time) and midnight local time, reloading a chart causes all data that's arrived since NY close to disappear. For example, a 5min chart refreshed at 2330 loses the six bars that have appeared since 2300. After midnight all the missing data reappears.

This is a new bug that wasn't in the previous version, but might be related to the old bug whereby daily charts behave differently between NY close and local midnight.

M Klein

Dear M Klein,
Thank you for the bug report. We haven’t seen it, but we will try to reproduce it. Could you tell me you local time zone?

Could you tell me your local time zone?

GMT+1 (Paris)

M Klein

Thank you! we will try to reproduce it.

Just one more thing for the moment.

The prior version of MCFX treated the week as lasting 5 days (Sunday open to Friday close), with any price movement on Sunday evening contributing to Monday's bar. The new version seems to allow for a 7 day week, with the result that price movement before 2200GMT on Sunday evening creates a new bar, containing little data. This affects the daily charts in particular. All daily charts now show "doji" bars for Sunday 30/12/2007 and Sunday 06/01/2008

N-day charts also now seem to include weekends as two trading days, so a 2-day chart has a doji bar for 05/01-06/01. This distorts the mid-term picture. It makes much more sense to treat the week as being composed of five days, as before.

M Klein

It's been a while since I posted about this; I wondered if there had been any progress. This is quite a big problem for the daily and n-day charts. I've attached some pictures showing the doji bars caused by weekend ticks, in most cases just before market open on Sunday evening. The problem dojis are circled in red.

M Klein

This is one of our high priority issues. As soon as we close the opened issues we will start working on this and 1hour/60 minutes bar discrepancies problems.

This is one of our high priority issues. As soon as we close the opened issues we will start working on this and 1hour/60 minutes bar discrepancies problems.

Thank you.

You are welcome!