From: Cook, G. <GW...@ma...> - 2004-08-17 12:46:54
|
Yves Mettier wrote: >> Garry, Yves, >>=20 >> I want to talk about RRD, however I have lost the original thread. >>=20 >> Let me explain my worries about using RRD. Please correct me where >> I am going wrong. >=20 > My comments will be my explanations. > Garry has his own opinion (I agree with him for most of what he said). >=20 >>=20 >> Round Robin Database gives a method of storing a fixed period of data >> which is stored locally and may be graphed by many available open >> source programs. This is a fast self-purging method liked by many >> people.=20 >>=20 >> But what are we proposing? >>=20 >> i) Using just RRD. >>=20 >> ii) Using RRD for storage of Binary data, and a DBMS for everything >> else.=20 >>=20 >> iii) Using PP as is, but also store data in RRD. >=20 > iv) Using RRD for storage of Binary data, and a DBMS for *all* >=20 > Well, propose everything ! > Some may only use the perfparse binary (not the CGI...) to > create RRD files and use them > with other tools. Some may need database storage. Some may > need both... >=20 > I my mind, there is only a --rrd-dump=3Drrdfile option to add > in perfparse.c and to support. > Those who want to graph the data from the rrdfile will use > their own tools, and maybe > some may send us a contribution ? At this point I'm not sure what to propose. I was originally proposing that we give users the ability to use ONLY RRD, if they wish. But I was afraid that this might no longer be possible, due to some of the features that have already been implemented.=20 Is this definitely the case? What is needed that cannot be stored in RRD? The groups I guess, and raw data, although I don't understand what part of that can't be stored, the strings I suppose. Yeah, I'm starting to see why we cannot use only RRD. Perhaps you guys can just throw in the option that Yves mentioned above and we can move on... >> I have seen some comments about using (i), just RRD. >>=20 >> If we do this, the product we will have effectively is not PerfParse. >> - There will be no storage of raw data. >=20 > For now, this is not a problem for me: I use only binary data. >=20 >> - The extra information we plan, like a range of WARN and CRITICAL >> values will not be supported. >=20 > This is a choice I can do for here. My need is that the extra > informations are parsed. I don't mind they are not stored. > And there is a hack we can do: store them as another value. > The rrdfile will grow > faster, with 5 values instead of 1 (val, warnmin, warnmax, > critmin, critmax) I'm not sure that you guys are correct here. The range values can all be stored, and it is not really a hack. I think you can store many values in an RRD, if that is your choice. =20 >> - The product will not be client/server, and therefore the whole of >> PP must be using on a single machine. >=20 > I need this. So I need the database (except if we make > another server/client system, but > this is not planned, and we have enough of them for now) I use all of my tools on a single machine anyway. In fact I prefer it that way. Should I ever go into business building network monitoring appliance servers, I wouldn't want to have to tell people that they must buy at least two servers. I understand that this just gives me more flexibility, and it is optional to have a second server, but if we were to pull a lot of the data out of MySQL and put it into RRD, I don't really think there would be a need for a second server, as I think it would perform much better. >> Effectivelly, the entire PP CGI is redendent and we should probably >> advise users to use one of the existing mature RRD viewers. > This will >> mean that PP is reduced to just the 'perfparse' program, which it's >> self is reduced to just storing binary values in an RRD database. >>=20 >> I don't think we want to do this. :) >=20 > I think that some want it. You would not have so many mails > about it otherwise. Yes. If you want users to use PP, RRD support is necessary. Those who want/need it are going to go over to Nagiosgraph otherwise. >> With option (ii), using RRD for the binary data. This also presents >> some problems:=20 >>=20 >> - Extra data, like range of WARN and CRIT cannot be supported in >> full. Although two lines for each can be stored to indicate range.=20 >> Messy!=20 You may be right, I'm not sure. It's been a long time since I looked at RRDTool in detail, although I use it every day for my MRTG graphs. However, I think you'd be surprised at what you can do with RRDTool. =20 > The users make the choice. >=20 >> - Half of PP is client/server, half is local. >=20 > This is good for data storage. Few data central, a lot of > data on each monitored machine. > If the CGI could ask each monitored machine for the data, it > would be better for data > storage : save the data with your usual backup procedure. It > would also be better for > network traffic: only the needed values go on the network. > Sadly, it is more complex to implement this because of the > client-server issue.=20 >=20 >> - Referential Integrity is hard to maintain using two different >> database types.=20 >> - Lots of re-coding to get the chart to read RRD data. >=20 > About this: probably like you, I fully disagree about the CGI > reading RRD data. > I agree with feeding RRD files, but exploitation of RRD files > has to be done with other > tools (that can be part of the perfparse project, one day, why not ?) >=20 >> Again, I wonder whether this is worth it, although there might be an >> argument.=20 >>=20 >>=20 >> My feeling is that we should provide an option the user can select >> when compiling:=20 >>=20 >> 1. Store data as RRD for their own use, or not. >> 2. Store data as PP format, or not. 3. Store data as PP format, and be able to dump to RRD at regular intervals, perhaps as data is purged from MySQL. > Yes :) > My own reflexion results in the same conclusion :) >=20 > Yves >>=20 >> In the second case, we are effectively providing a method to import >> data into RRD, and the users chooses not to use any of the other >> features we provide.=20 >>=20 >> As always, I look forward to your comments. >>=20 >> Ben Garry W. Cook, CCNA Network Infrastructure Manager MACTEC, Inc. - http://www.mactec.com/ 303.308.6228 (Office) - 720.220.1862 (Mobile) |