User Benchmarks

[erikatcuse]

I was looking at the User Benchmarks Wiki http://lime-technology.com/wiki/index.php?title=User_Benchmarks and it hasn't had any updates in a while.

 

If you want to post your how many drive you have, what type (IDE, SATA, MIXED), Smallest drive size, Largest drive size, Motherboard model, HD controller cards (or motherboard chipset if you know it) and your Average Parity Check Speeds

 

to this post.  I'll update the wiki

 

 

[RobJ]

Unfortunately, when we started the User Benchmarks wiki page, we did not give clear instruction as to how to obtain the true Average Parity Check Speed, so the values have been derived in a number of ways, and aren't very reliable at all.  We should probably have indicated that the correct way is to obtain the actual number from the syslog, in KB/s, not in the user-derived MB/s number that it currently uses.  As it is, some of the numbers appear to be the highest number they saw while it was running, or an average of some of the numbers they saw, usually near the beginning of the run.  We could attempt to revise the table, but I'm not sure it is worth it.  I do hope to add some notes to the section, that will help to put the numbers in perspective, as well as note some of the factors that affect the speed of parity checking.  I don't believe some of the users who have helpfully contributed to the page, fully understood what made the numbers useful, helpful.  As one example, a user might add several drives to their system, then update the Benchmarks with the new drive count and changed parity check speed.  It isn't nearly as useful to know what a particular user's current speed is, as it is to know how the change in their hardware affected the speed.  It would have been more helpful to see the old data AND the new, by adding a modified copy of their previous line in the Benchmarks.  As another example, it might be useful to see the effect of upgrading the parity drive, by showing the parity check speed with the old drive as well as with the new drive.  In other words, please don't update a line, add another just below it, so we can compare.

 

I'm also not sure how useful the last 2 columns are, any more.  I doubt if many users will even understand what the current entries in the NCQ column really mean.  I don't know what to recommend here.  The only way I can see the unRAID version column being useful, is if a user found a significant difference in parity check speed between different versions, using identical hardware.

 

I have added 2 new sections to the User Benchmarks page, the Hard Drive Read Speeds section and the Preclear Times section.  Both are tentative, I can easily remove them if there is no real or sustained interest.  The column choices are also tentative, as others may have ideas that would be useful.  There is a Notes column for miscellaneous stuff.  The first was to provide a place to collect drive speeds for the many drives that unRAID users have, or are considering purchasing.  The second is to help users know how long preclearing a particular drive is likely to take, not a very high priority item, but perhaps helpful.  There are already a number of threads with numbers that could be transferred to these sections, by anyone interested in this project, or I'll get around to it at some point.

 

Concerning the Hard Drive Read Speeds section, I'd be interested in hearing knowledgeable opinions as to the methods suggested for measuring these speeds, as well as other simple methods.  I suggested both the hdparm -tT and the dd copy methods (as used here), would like to know how close their results are.  I could add a column for measurement method used, unless these methods are essentially equivalent for our purposes.