Today I did a thing, that I postponed many days. I’ve finally moved Pytrend environment to Linux.
In this short demo I am showing you that Pytrend now can process HPE 3PAR stats dumps and build various system level graphs, such as:
- IOPS over time (read/write/total)
- MB/sec over time (read/write/total)
- Latency over time (read/write/average)
- IO Size over time (read/write/average)
- CPU Utilization over time (system/user)
All above stats are directly parsed from dump into the graphs.
Though, there is another one, which I’ve added myself, because it is very useful to see how latency depends on IOPS, to see whether system resources (disks or controllers)are close to saturation.
In this little demo I show you two improvements for storage stats processing.
One of them is small enhancement of IBM DS3000/DS5000 series statistics. Originally, DS-series do not provide average IO Size in their service dump, so I have taught Pytrend to recognize bandwidth (KB/sec or MB/sec) and IOPS data to correctly calculate average IO size.
Another small enhancement for graphs now shows Average and 95-percentile for each plotted value. It is required in most situation for determining correct workload profile. This feature is available for all storage types.
This is aimed at partners and Dell people…..
I’m often asked to analyse the performance of 3rd party arrays before I build a solution. We have many tools to analyse behavior, but sometimes you have to analyse the raw data from the 3rd party array, rather than use host-based tools, especially if there are hundreds of servers connected to the same array.