Fog computing brings the functionality of the cloud near the edge of the network with the help of fog devices/micro data centers ($mdcs$). Job scheduling in such systems is a complex problem due to the hierarchical and geo-distributed nature of fog devices. We propose two fog scheduling algorithms, named $FiFSA$ (Hierarchical $Fi$rst $F$og $S$cheduling $A$lgorithm) and $EFSA$ ( Hierarchical $E$lected $F$og $S$cheduling $A$lgorithm). We consider a hierarchical model of fog devices, where the computation power of fog devices present in higher tiers is greater than those present in lower tiers. However, the higher tier fog devices are located at greater physical distance from data generation sources as compared to lower tier fog devices. Jobs with varying granularity and cpu requirements have been considered. In general, jobs with modest cpu requirements are scheduled on lower tier fog devices, and jobs with larger cpu requirements are scheduled on higher tier fog devices or the cloud data center $(cdc)$. The performance of $FiFSA$ and $EFSA$ has been evaluated using a real life workload trace on various simulated fog hierarchies as well as on a prototype testbed. Employing $FiFSA$ offers an average improvement of 27% and 57.9% in total completion time and an improvement of 32% and 61% in cost as compared to Longest Time First ($LTF$) and cloud-only ($cdc-only$) scheduling algorithms, respectively. Employing $EFSA$ offers an average improvement of 48% and 70% in total completion time and an improvement of 52% and 72% in cost as compared to $LTF$ and $cdc-only$ respectively.