mdcallag / linkbench Goto Github PK

Summaries are printed every N inserts or queries. I want an option to do this every N seconds.

From a brave user...

1. Suppose I load 'x' number of nodes. I expect to see 'x' number of rows in nodetable after load phase. With the pervious old version(2011), the number of rows were as expected. But, with the new repository, the number of rows after load phase is '0'. The linktable shows the correct number of rows but not the 'nodetable'. Here's an example

LOAD PHASE COMPLETED.  Loaded 4 nodes (Expected 4). Loaded 961966 links (240491.50 links per node).  Took 811.6 seconds.  Links/second = 1185

Load Phase completed....

Number of rows in linktable: count  
--------
 961966

Number of rows in counttable: count 
-------
     4

Number of rows in nodetable: count 
-------
     0

2. When I use the username 'linkbench' as suggested in the README, it for some reason looks for database with that name. Not sure why. So, it only works when the username is same the database name which is 'linkdb'.

p99 rates for inserts and queries per second are derived from the rates of one thread using a shell script. It would be nice to compute these directly in Python across all threads.

Hi Mark

I realize this is not very actively maintained, but recording this here while I'm deep diving into Linkbench a little. Comments on options for fixing this are welcome.

For development purposes, and maybe other reasons, I want to run quicker runs of Linkbench. I can make load phase quicker by loading less data. If I set maxid1 = 10001 I get a smaller nodetable and counttable. However, the load phase then continues to insert 44M records to linktable. This also means there will be on average thousands of links per node, which is probably not a good model of reality.

It appears the default links generator nlinks_func = real isn't configurable at all. Should it be? Maybe is should automatically adapt to maxid1 so that it always generates about 4 links per node?

I can workaround this issue by using another nlinks_func:

nlinks_func = com.facebook.LinkBench.distributions.ZipfDistribution
nlinks_shape = 1.5
nlinks_mean = 2