[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Feed Performances Under Stress!



PureBytes Links

Trading Reference Links

As some have noted, today was a day which really tested the ability of the
various data feeds to deliver under stress.  Performance of one was
amazingly good, while performance of another was amazingly poor.

If anyone missed the explanatory preamble which I posted a few days ago and
would like another copy, please e-mail me privately and I'll send it.

Regards,
Carroll



Feed analysis for 08/10/1999
Lag is difference between CME T&S time & collected time stamp.
Gap length is in seconds.

ESignal - TS2000

End      Lag      Tick Counts          Gaps
Time   Seconds  CME  Feed   Ratio   Lnth Ticks
10:00     1     340   334   0.982    25    1
10:30     0     357   350   0.980    28    1
11:00     1     288   282   0.979    28    1
11:30     1     358   352   0.983    22    1
12:00     0     313   312   0.997    10    1
12:30     0     329   327   0.994    37    1
13:00     0     399   396   0.992    18    1
13:30     0     275   273   0.993    25    1
14:00     1     305   300   0.984    24    1
14:30     1     314   309   0.984    25    1
15:00     0     313   305   0.974    19    1
15:30     1     329   327   0.994    22    1
16:00   -14     277   196   0.708   202   26


ESignal - MetaStock

End      Lag      Tick Counts          Gaps
Time   Seconds  CME  Feed   Ratio   Lnth Ticks
10:00   -10     342   333   0.974    25    1
10:30    -9     357   350   0.980    28    1
11:00   -10     287   279   0.972    28    1
11:30   -10     360   354   0.983    22    1
12:00   -10     311   310   0.997    10    1
12:30   -10     330   328   0.994    37    1
13:00   -10     397   393   0.990    18    1
13:30   -10     275   273   0.993    25    1
14:00   -10     306   300   0.980    24    1
14:30   -10     313   307   0.981    25    1
15:00   -11     315   304   0.965    36    3
15:30    -9     329   296   0.900   110   22
16:00   -10     277   269   0.971    32    1


Signal/cable

End      Lag      Tick Counts          Gaps
Time   Seconds  CME  Feed   Ratio   Lnth Ticks
07:00    31     347    63   0.182   190   30
07:32    11     350    78   0.223   189   46
08:00     4     288   204   0.708   195   17
08:30     3     356   313   0.879    34    2
09:00     1     315   301   0.956    35    2
09:30     0     329   326   0.991    37    1
10:00     1     399   395   0.990    18    1
10:30     0     275   269   0.978    25    1
11:00     1     305   298   0.977    27    1
11:30     3     313   292   0.933    29    1
12:00     4     314   232   0.739    41    2
12:30     3     329   260   0.790    51    2
13:00     3     276   184   0.667    68    7


BMI/cable - TS4

End      Lag      Tick Counts          Gaps
Time   Seconds  CME  Feed   Ratio   Lnth Ticks
10:00     4     338   338   1.000     0    0
10:30     4     356   356   1.000     0    0
11:00     4     290   290   1.000     0    0
11:30     4     356   356   1.000     0    0
12:00     4     316   316   1.000     0    0
12:30     4     329   329   1.000     0    0
13:00     4     399   399   1.000     0    0
13:30     4     275   275   1.000     0    0
14:00     4     303   303   1.000     0    0
14:30     4     315   315   1.000     0    0
15:00     4     314   314   1.000     0    0
15:30     4     329   329   1.000     0    0
16:00     3     277   277   1.000     0    0
16:21     4     145   143   0.986    16    1


Signal - all feeds:

   One puzzling occurrence is that there almost always is exactly one
   bad tick in each day's Signal data and it appears identically in
   all three of the above feeds.  Very, very few days occur during
   which this single bad tick fails to appear.  The error is not
   usually large - today's is the largest I've seen:  at 12:18:41 NY
   time, 1286.70 was reported by Signal as 1287.70.  This error is so
   reliable that it's almost as if it's being injected deliberately! 
  
   Today is the first time I've seen a corresponding bad tick in the
   BMI/cable feed. 
  
  
Signal/cable:
  
   Performance was completely worthless during the first hour; 80% of
   the ticks were missed!  And the few which WERE received were 31
   seconds late.  It finally struggled up to 99% (1% loss) by midday
   but then declined again going into the afternoon.  In the last
   half hour it was missing 1/3 of all ticks. 
  
  
ESignal:
  
   ESig was considerably better - until the afternoon.  One
   significant observation is that except for some large missing
   blocks of ticks late in the day, almost ALL of the scattered
   individual missing ticks appeared identically in both of the two
   data collections.  This shows that ESig failed to send them;  they
   were not lost within the local, collecting system. 
  
   Although they did not occur at the same time for the two
   subscribers, both suffered losses of large blocks of ticks in the
   late afternoon.  The TS2000 user's data is missing ticks from
   14:35:10 (NY time) to 14:36:22, from 14:41:03 to 14:43:41, and
   from 14:54:40 to 14:56:49.  The MetaStock user's data has a gap
   from 14:09:02 to 14:11:04. 
  
   One example of the effect of missing even a few ticks is that
   three ticks missed at 13:49:21 missed the local (5-minute) high of
   1282.00.

  
BMI/cable:
  
   What can I say?!  I don't think I've ever before seen a perfect
   score - NOT A SINGLE TICK MISSED and the delay is a constant and
   reasonable 4 seconds! 

   Today (8/10) was clearly a day which greatly stressed the feeds -
   for example, I've never before seen the Signal/cable feed so badly
   overloaded as it was today.  So, for the BMI/cable feed to have
   performed as it did is nothing short of remarkable!  (The one
   "tick" missing at the end is, I think, a program glitch, because
   the files match perfectly except for two bad ticks, one at
   10:19:11 (NY time) and one at 12:18:41.