The mean latency of our experiments is 0.605s, with a maximum value of 0.834s. This delay has two main components: the operating system polling frequency and the time it takes for transceivers to complete the fault recovery process, before ports come up and data transfer can start or resume. PicOS polls the transceivers every 250ms; furthermore it only supports duplex auto negotiation mode for SFP. On the other hand, the SFP vendor informed that the initialization delay varies across different brands, but it should be consistent with the SFF8431 guidelines for SFP+ DWDM. After a fault is detected in the link, in less than the maximum value of t start up plus t reset , the optical transmitter reset the laser circuits and disable the tx fault flag.Moreover, the Polatis optical switch takes 25ms at most to complete the optical reconfiguration, as listed in the technical documentation. With our MBB approach we first updated the flows in the EPS tables sending instructions from the controller, then we provisioned the new link with the optical switch, and finally we updated the flow again in the EPS to force the traffic to pass through the new path. In a single stream of data, RTT grows up to hundreds of millisecond in the OST reconfiguration, but does not grow larger than 2.5 ms with the MBB mode. On average, the packet loss decreases from 2.8% in OST to 0.93% in MBB. Link unavailability due to RTO events went from 598 ms to 121 ms, 80% less. With MBB, throughput drops 0.6 Gbps, grow rack compared to the drop by 6 Gbps in OST. Bandwidth steering was also improved with our MBB approach. This scenario was recreated with two data streams at 20 Gbps in total, forcing a bottleneck of 10 Gbps in the testbed.
Comparing performance during reconfiguration, packet loss went from 1.28% with OST to 0.44% MBB. Link unavailability decreased from 727 ms to 185 ms, 74% less. RTT grows up to hundreds of millisecond in OST, but is not larger than 2.5 ms like in the single data stream scenario. We did not achieve hitless reconfiguration, that is, 0% packet loss, because there still was a link unavailability that generated packet drops. However, we decreased packet loss with our MBB approach, getting closer to the ideal goal. The first step to reach zero packet loss would be to decrease the transceiver locking and switch polling latency, so we do not trigger the TCP RTO timer. Other options would be to play with different RTO values, or to replace TCP with other protocols. In the future, our testbed will be used for deploying optical networks with technologies other than MEMS, and running experiments in different areas such as machine learning and heterogeneous computing. The modular design allows modifying blocks of software or hardware described in chapters of architecture and infrastructure, without affecting other elements of the system. Agriculture in the United States has undergone massive consolidation over the past 50 years and the same is true in California. Several economic and market factors have contributed to farm consolidation, but new regulations on agriculture have also played a role . Compliance costs associated with increased regulatory burdens can decrease producer profits and limit market entry . Small producers may be particularly harmed by the need to achieve compliance, as economies of scale provide larger producers an advantage . Small firms may lack sufficient capital to change production methods to comply with regulations, or even to manage the burdens associated with reporting.
The cannabis industry has historically resisted widespread farm consolidation, perhaps due to its status as an unregulated, and illicit or semi-licit, activity. While the amount of cannabis produced in California is substantial , evidence from 2016 suggests that most outdoor cannabis was then produced on farms smaller than one acre . When Proposition 64 legalized non-medicinal cannabis in 2016, its size provisions explicitly acknowledged the state’s desire to see cannabis farms remain small . Initial regulations limited each permit to an area no greater than one acre and limited each entity to only one permit. Federal laws against cannabis have also encouraged small farms: Farmers with more than 99 plants potentially face federal minimum sentences of five years in prison . Local permitting may also favor smaller producers. Each jurisdiction in California can create its own permitting system, and possessing a local permit is a condition for obtaining a state permit. Most local jurisdictions place limitations on field sizes, and these limitations can encourage small-scale farming. While local permits may provide an avenue for local governments to protect small farmers , they also add another layer of regulation, potentially increasing entry costs. Beginning with California’s first attempt to implement a comprehensive regulatory system for the cultivation and distribution of legal cannabis, through the 2015 passage of the Medical Marijuana Regulation and Safety Act, stakeholders have expressed concerns that the permitting process privileges large farms over small. MacEwan et al. calculate that, due to the nature of regulatory costs, the type of small cannabis farmer prevalent in Northern California is the “least likely to participate in the regulated market.” Yet to date, empirical evidence on cannabis producers’ engagement with the formal market under the new regulatory framework has been lacking. In particular, there is a large evidence gap about the types of farms that participate in the regulated market and those that do not.
We remedy that gap by combining information about farmers who have started the permit application process with a unique dataset of cannabis farms in Humboldt County in 2012 and 2016. Humboldt County is one of the largest cannabis producing regions in California and perhaps the world. Cannabis farming began there in the early 1960s, with rapid expansion following in the 1970s, and cannabis has been among the most valuable crops in the county at least since a proposition legalizing medical cannabis was approved by voters in 1996 . Recent studies suggest that at least 5,000 cannabis farms operate in Humboldt County . In the lead-up to the enactment of regulated cultivation of cannabis — which began for the medicinal market in 2016 and for the adult-use market in 2018 — the region experienced a cannabis boom, with the number of plants under cultivation increasing by 150% between 2012 and 2016 . This time of massive cannabis expansion is often referred to locally as the “green rush.” To track both permitted and unpermitted cannabis growers, we used data created by Butsic et al. . In their study, Butsic et al. hand-digitized cannabis farms using very high resolution satellite imagery. Cannabis production was measured in both 2012 and in 2016. Outdoor plants were counted and the number of plants inside greenhouses was estimated based on greenhouse size. Of the 1,724 farms in the dataset, 942 started producing cannabis between 2012 and 2016 and 782 produced at least some positive amount in both 2012 and 2016 . For permit data, we used publicly available data from the Humboldt County Planning Department, compiled from applications for commercial cannabis cultivation permits . We were able to combine the farm location data with the permit data based on the unique parcel identification that existed in both datasets. In total, hydroponic rack applications were received for cultivation on 1,945 unique parcels. Of these, 533 were located within our study area . We also include data describing farm/parcel characteristics. Locational variables such as distance to public roads and cities are used to proxy for transportation cost, while distances to endangered and threatened fish species habitat proxy for the environmental sensitivity of a site. Distance to ocean provides a summary measure of the coastal environment of the farm. Biophysical characteristics such as slope and presence of prime agricultural soils are used to describe the growing conditions of a site, while zoning designations are used to identify areas where growing cannabis is allowed . We also determined if a timber harvest plan had been associated with a parcel at any point since 1997. The overall aim of our empirical analysis is to describe the type of cannabis farms likely to apply for a permit.
To do this we use a twofold approach. First, we compare farms that applied for a permit and farms that did not in terms of the means of their farm and parcel characteristics. We use a simple two-tailed test to determine if the univariate mean differences between these groups are statistically significant. We focus on differences in farm size , farm-size expansion during the “green rush” period and tenure of the farm. In a second step we estimate models of application decisions using multivariate regressions, which allow us to isolate the impact of each characteristic while controlling for variation in others . We implement two such models. Our main specification is a probit model in which the binary dependent variable is equal to 1 if a permit application was submitted for parcel i. The size of the farm is included with a quadratic specification and the other parcel and farm characteristics enter the model linearly as independent variables. We use the probit model to estimate the marginal contribution of each of these variables to the likelihood that a parcel applies for a permit.The average farm size in 2016 was 432 plants, with a median of 263 plants, a minimum of 14 and a maximum of 12,901 . Over 90% of farms produced fewer than 1,000 plants and fewer than 2% produced more than 2,000. Examining permit application rates by farm size reveals a distinct size gradient , as application rates increase substantially over farm-size categories. This pattern holds for both existing and new farms, but the rise is much sharper for the latter. Approximately 10% of small new farms apply for a permit, but rates jump to 61% and 50%, respectively, for the largest farm size groupings. We found a significant difference in size between farms that applied for a cannabis permit in 2016 relative to those that did not apply . The trend according to which larger farms applied for permits at higher rates held true regardless of production type . The size differences are proportionally similar for both greenhouse and outdoor plants, so we do not find evidence that the relationship between farm size and permit application is solely driven by production method. Our regression models confirm that this result is robust to controlling for other covariates. In all our regression specifications, the coefficient on the total number of plants in 2016 is positive and statistically significant at the 1% level. The effect size of the number of plants indicates that, controlling for parcel characteristics, an increase of 100 plants increases the probability of applying for a permit by 2.4% , with the slope of the relationship declining for extremely large farms . The overall marginal effect is similar for existing and new farms, , though the declining marginal effect for very large farms is driven by new farms , and is robust to the inclusion of watershed fixed effects . The pattern also holds for size in 2012. Restricting the sample to existing farms, an increase of 100 plants in 2012 increases the probability of application by 3.1%.We first categorize growth of existing farms according to the proportionate change in plants produced between 2012 and 2016. The “declining production” group consists of farms that shrank by more than 5% ; “minimal change” farms experienced between −5% and 5% growth ; “moderate growth” farms grew between 5% and 50% and “high growth” farms grew by more than 50% . Within the sample of existing farms, there is a clear gradient of application rates with respect to growth between 2012 and 2016 . The farms least likely to apply are those that declined in size, followed by those with minimal growth. Application rates for existing farms that grew moderately jump to over 40%, with high growth farms the most likely to apply. Note that across all expansion rates for existing farms, application rates are significantly higher than the average rate for new farms.