EBS Metrics

Collected

Fully Qualified Name (FQN)	Description	Units	BASE	CORR	UTIL	Related Global Policies
netuitive.aws.ebs.averagelatency	This metric expresses the average latency per operation for this EBSinstance. This metric is useful for monitoring EBS performance andquality of service.Computation:(Total Read Time + Total Write Time) / (Read Ops +Write Ops)	seconds	yes	yes	no	Elevated Queue Length Differential with Elevated Latency
netuitive.aws.ebs.iops	This metric expresses the total number of either read or writeoperations per second for this EBS instance. This metric is useful formonitoring EBS I/O activity.Computation: Total Ops / 300	operations/second	yes	yes	no
netuitive.aws.ebs.busytimeiops	This metric expresses the number of operations per second measured only over the time that the disk is actually busy. This can be useful indetermining the IOPS being achieved during bursts.Computation: Total Ops / (300 – floor(Idle Time))	operations/second	yes	no	no

| netuitive.aws.ebs.queuelengthdifferential | This metric is measuring the difference between the actual queue lengthand the “ideal” queue length. The ideal queue length is based onAmazon’s rule of thumb that for every 200 IOPS you should have a queue length of 1. In theory, a well-optimized volume should have a queue length differential that tends to hover around 0. In practice, we have seen volumes with extremely low latency (< 0.0001) have queue length differentials that are higher than 0; presumably this is because the latency is much lower than Amazon is assuming for their rule of thumb. Even in these cases, the differential is a pretty steady number; hence an upper deviation in the differential would tend to indicate that the disk is not keeping up.Computation: Queue Length – (IOPS / 200) | difference | yes | no | no | Elevated Queue Length Differential with Elevated Latency | | netuitive.aws.ebs.iopsutilization | This metric compares the current IOPS to the provisioned IOPS for thevolume in order to determine how much of the provisioned capacity isbeing used.Computation:min(100, (attribute[IOPS] == NULL ? data[IOPS] / 300 : data[IOPS / attribute[IOPS]]) *100) | percent | yes | yes | yes | |

Computed

Fully Qualified Name (FQN)	Description	Units	BASE	CORR	UTIL	Related Global Policies
netuitive.aws.ebs.totalops	This metric expresses the total number of read and write operations against this EBS instance. This metric is useful for monitoring EBS I/Oactivity.Computation: Read Ops + Write Ops	operations	yes	no	no
netuitive.aws.ebs.totalbytes	This metric expresses the total number of bytes read and written from/to this EBS instance. This metric is useful for monitoring EBS I/Oactivity. Computation: Read Bytes + Write Bytes	bytes	yes	no	no
netuitive.aws.ebs.averagelatency	This metric expresses the average latency per operation for this EBSinstance. This metric is useful for monitoring EBS performance andquality of service.Computation:(Total Read Time + Total Write Time) / (Read Ops +Write Ops)	seconds	yes	yes	no	Elevated Queue Length Differential with Elevated Latency
netuitive.aws.ebs.readbytespersec	This metric expresses the bytes read per second from this EBS instance.This metric is useful for monitoring EBS read activity.Computation: Read Bytes / 300	bytes/second	yes	no	no
netuitive.aws.ebs.writebytespersec	This metric expresses the bytes written per second to this EBS instance.This metric is useful for monitoring EBS write activity.Computation: Write Bytes / 300	bytes/second	yes	no	no
netuitive.aws.ebs.totalbytespersec	This metric expresses the total number of bytes either read from orwritten to this EBS instance. This metric is useful for monitoringoverall EBS I/O activity.Computation: Read Bytes Per Second + Write Bytes Per Second	bytes/second	yes	yes	no
netuitive.aws.ebs.averagereadlatency	This metric expresses the average latency per read operation for thisEBS instance. This metric is useful for monitoring EBS performance andquality of service.Computation: Total Read Time / Read Ops	seconds	yes	no	no
netuitive.aws.ebs.averagewritelatency	This metric expresses the average latency per write operation for thisEBS instance. This metric is useful for monitoring EBS performance andquality of service.Computation: Total Write Time / Write Ops	seconds	yes	no	no
netuitive.aws.ebs.readopspersec	This metric expresses the number of read operations per second for thisEBS instance. This metric is useful for monitoring EBS read activity.Computation: Read Ops / 300	operations/second	yes	no	no
netuitive.aws.ebs.writeopspersec	This metric expresses the number of write operations per second for thisEBS instance. This metric is useful for monitoring EBS write activity.Computation: Write Ops / 300	operations/second	yes	no	no
netuitive.aws.ebs.iops	This metric expresses the total number of either read or writeoperations per second for this EBS instance. This metric is useful formonitoring EBS I/O activity.Computation: Total Ops / 300	operations/second	yes	yes	no
netuitive.aws.ebs.busytimeiops	This metric expresses the number of operations per second measured only over the time that the disk is actually busy. This can be useful indetermining the IOPS being achieved during bursts.Computation: Total Ops / (300 – floor(Idle Time))	operations/second	yes	no	no
netuitive.aws.ebs.busytimebytespersecond	This metric expresses the number of bytes per second read and written,measured only over the time that the disk is actually busy. This can beuseful in determining the maximum throughput being achieved duringbursts.Computation: Total Bytes / (300 – floor(Idle Time))	bytes/second	yes	no	no

| netuitive.aws.ebs.queuelengthdifferential | This metric is measuring the difference between the actual queue lengthand the “ideal” queue length. The ideal queue length is based onAmazon’s rule of thumb that for every 200 IOPS you should have a queue length of 1. In theory, a well-optimized volume should have a queue length differential that tends to hover around 0. In practice, we have seen volumes with extremely low latency (< 0.0001) have queue length differentials that are higher than 0; presumably this is because the latency is much lower than Amazon is assuming for their rule of thumb. Even in these cases, the differential is a pretty steady number; hence an upper deviation in the differential would tend to indicate that the disk is not keeping up.Computation: Queue Length – (IOPS / 200) | difference | yes | no | no | Elevated Queue Length Differential with Elevated Latency | | netuitive.aws.ebs.iopsutilization | This metric compares the current IOPS to the provisioned IOPS for thevolume in order to determine how much of the provisioned capacity isbeing used.Computation:min(100, (attribute[IOPS] == NULL ? data[IOPS] / 300 : data[IOPS / attribute[IOPS]]) *100) | percent | yes | no | no | |