Toad World® Forums

TPC-C Delivary transaction concerns


#1

What ever TPC-C benchmarks I run, when looking at the transaction and response times for the delivary transaction, I allways get a low time for the first user load, an extreamly high time for the second userload tested and then subsiquent userloads are all consistent… its very strange , any ideas?

ie if the user loads are 1,2,3,4,5 … for one user the tratinsaction time is 5 m/s there is the huge spike on user load 2 of 300m/s then every subsiquent user load tested is steady at 80 m/s

if the user loads are 1,5,10,15,20 … for one user the tratinsaction time is 5 m/s there is the huge spike on user load 5 of 300m/s then every subsiquent user load tested is steady at 80 m/s

if the user loads are 1,10,20,30… for one user the tratinsaction time is 5 m/s there is the huge spike on user load 10 of 300m/s then every subsiquent user load tested is steady at 80 m/s

allways on the delivery transaction …

any of an explanation for this?


#2

Not sure why you are seeing the results you are seeing without getting more data and doing some performance analysis. Here is what the Delivery transactions does;

Delivery (4% of Mix)
Processing a batch of 10 new (not yet delivered) orders. Each order is processed (delivered) in full within the scope of a read-write database transaction
Warehouse ID is the one assigned the virtual user
Carrier ID is random 1-10
Delivery date is the SUT system date

I have looked at several of my test runs and only see one out of the 10 or so that I have looked at so far that had the Delivery transaction go up on the second iteration and then back down for the rest.

How many executions are you getting of the Delivery transaction for the different user loads? If you are not getting enough samples to get a good statistical sampling, what happens if you increase the Sampling time of an iteration?