AI Zone Admin Forum Add your forum

NEWS: Chatbots.org survey on 3000 US and UK consumers shows it is time for chatbot integration in customer service!read more..

AWS Performance test. General Purpose SSD versus Provisioned IOPS SSD
 
 

If you are using AWS, you might be wondering about the performance difference between
general-purpose SSD versus provisioned SSD.
Well,  Provisioned SSD is a lot more expensive and it should be faster.

To test this, I ran a simple regression test, 1000 volleys against each
AWS gs2 general purpose SSD
versus
AWS Provisioned IOPS SSD io1

Everything else was the same. Same code, same machine profiles.  It was run using the command console, to eliminate the network impact.

I experienced a 2% improvement for the provisioned SSD.
Based on this test, I do not think it is worth the ssd upgrade if you are using aws for CS.

GS2: Sourcefile Time used 43341 ms for 1214 sentences 5631 tokens.
IO1: Sourcefile Time used 42485 ms for 1214 sentences 5631 tokens

As an aside, I also tested CS 9 versus CS 9.6 and it is about the same.

 

 
  [ # 1 ]

As one might expect, because CS is cpu bound and is not IO intensive.
Execution speeds of various versions of CS rarely vary much as the underlying algorithms are the same (mostly).
If you want to globally impact execution speeds, you would dedicate more dictionary buckets to speed up hash lookup. But still I would not expect huge improvements. And there is compiling with more optimations, but I have seen issues with that at times, so I dont trust max optimization. We use -g (linux) at work because I have had mysterious failures of pattern detection under higher optimizations.

 

 
  [ # 2 ]

hmm. I think I have to revisit my buffer settings.
I reran the 1000 volume test with :show stats, and I think I may have to change my buffers
this is what I am seeing now
  maxrules=85021 maxdict=574 maxfact=303 maxtext=231804

It looks like it is going pretty deep for a response when the maxrule number jumps.

I am not totally clear on how to change the buffers. If I need to.

CommandLine: /home/ec2-user/ChatScript
  Timer=3000
  nodebug
  Fork=5
  Noserverlog
  repeatLimit=5
  local

ChatScript EVSERVER Version 9.62 pid: 29126 64 bit LINUX compiled Sep 1 2019 18                                                                                                         :45:58 host=local
Params:  dict:2097151 fact:800000 text:100000kb hash:215127
      buffer:80x80kb cache:1x5000kb userfacts:100 outputlimit:80000 loglimit                                                                                                         :80000
WordNet: dict=201008 fact=86302 heap=11921136 Aug19’19-08:42:06
Build0:  dict=277840 fact=392882 heap=5045696 Compiled:Sep01’19-19:58:28 by ve                                                                                                         rsion 9.62 “0”
Build1:  dict=7096 fact=3632 heap=9604148 Compiled:Sep01’19-21:50:13 by versio                                                                                                         n 9.62 “Harry”
Used 103MB: dict 485,944 (42763kb) hashdepths 31674.53467.49680.34960.21420.1188                                                                                                         7.6113.3203.1456.702.328.146.53.20.11.7.1. words 485944 unusedbuckets 31674 fact                                                                                                         482,816 (23175kb) heap 26570kb
        buffer (6400kb) cache (5000kb)
Free 88MB: dict 1,611,206 fact 317,184 stack/heap 73,429KB

  Postgres enabled.

 

 
  login or register to react