Thanks:  0
Likes:  0
Dislikes:  0

# Thread: 50th Percentile time to last byte

1. ## 50th Percentile time to last byte

Hi, there,

I am always confused with "50th Percentile time to last byte".
If I have a result for a certain request like:
Times 1 2 3 4 5 6 7 8 9 10
TTLP 1 1 1 1 2 2 2 3 3 4

Then what is the 50th TTLP? Could somebody give me the formula?

Thanks,
Jinse

2. ## Re: 50th Percentile time to last byte

What tool is that from?

50th percentile means that half the values are larger than that number and half are smaller.

I have no idea what TTLP stands for.

3. ## Re: 50th Percentile time to last byte

Sorry, not TTLP, it should be TTLB.

4. ## Re: 50th Percentile time to last byte

Ok, Time To Las Byte - the way MOST tools have their DEFAULTS set up (note that there can be huge variants by tool and configuration) is that it times "response" as the time from when the first byte of a request leaves the client computer until the time that the last byte of the response is received by the client computer.

5. ## Re: 50th Percentile time to last byte

Thanks, Scott.

One more question though:

Which data you think more important, Average TTLB or 50th TTLB?

I want to fill a matrix, but don't know which one is better.

6. ## Re: 50th Percentile time to last byte

I always report on either the 90th or 95th (depending on sample size). 90th implies that "9 out of 10 people will experience this time or less". To me that is a much more meaningful number than either the average or the 50th.

In an evenly distributed sample, average and 50th would be identical. If you have to use one of the two I would definately use 50th.

7. ## Re: 50th Percentile time to last byte

In statistics there are 3 points of central tendency; the mean, mode and median. They all have their pros and cons. For performance measurements, the median is a better number to use than the mean (average). Averages are swayed by high and low numbers. The median (50%) is not.
Anyway, the 50% (median) and the 95% are both important. Typically the 95% is deemed worst case scenario. Any number you use though, stick with it for comparison between test runs. Scott, you mentioned in a perfect (bell curve) distribution, the median and mean would be identical. This is true, but never happens. Usually response time test results are positively skewed where the median is less than the mean. Many times you'll get multi-modal distributions that are due to the data used in the test as well. I can probably write about this forever, but getting back to the original question, we all agree that the 50% is the correct and better number to use.

8. ## Re: 50th Percentile time to last byte

JHarris,

I hope I didn't imply that you could ever expect the average (mean) and 50th percentile (median) to BE identical in a performance test. I was just trying to point out that many people get them confused because in some circumstances they happen to be the same.

9. ## Re: 50th Percentile time to last byte

Originally posted by RSBarber:
In an evenly distributed sample, average and 50th would be identical.
<font size="2" face="Verdana, Arial, Helvetica">Real life example (numbers were a little bit rounded):
95% of all responses have about the same response time.
5% of the responses have 2000% response time of the first group (20 times more).

In this case:
Median = Min
Average = 2 x Min
94th percentile = Min
96th percentile = 2 x Min

In this situation the best parameter would be a percent of the responses above SLA response time.

10. ## Re: 50th Percentile time to last byte

That is very similar to the example I use to explain those numbers, but I think yours is even easier to follow. That's excellent, Thanks!

#### Posting Permissions

• You may not post new threads
• You may not post replies
• You may not post attachments
• You may not edit your posts
•
Search Engine Optimisation provided by DragonByte SEO v2.0.36 (Pro) - vBulletin Mods & Addons Copyright © 2016 DragonByte Technologies Ltd.