Lightweight Internet Traffic Classification - A Subject Based Solution with Word Embeddings

Internet traffic classification is a relevant and mature research field, anyway of growing importance and with still open technical challenges, also due to the pervasive presence of Internet-connected devices into everyday life. We claim the need for innovative traffic classification solutions capable of being lightweight, of adopting a domain-based approach, of not only concentrating on application-level protocol categorization but also classifying Internet traffic by subject. To this purpose, this paper originally proposes a classification solution that leverages domain name information extracted from IPFIX summaries, DNS logs, and DHCP leases, with the possibility to be applied to any kind of traffic. Our proposed solution is based on an extension of Word2vec unsupervised learning techniques running on a specialized Apache Spark cluster. In particular, learning techniques are leveraged to generate word-embeddings from a mixed dataset composed by domain names and natural language corpuses in a lightweight way and with general applicability. The paper also reports lessons learnt from our implementation and deployment experience that demonstrates that our solution can process 5500 IPFIX summaries per second on an Apache Spark cluster with 1 slave instance in Amazon EC2 at a cost of $ 3860 year. Reported experimental results about Precision, Recall, F-Measure, Accuracy, and Cohen's Kappa show the feasibility and effectiveness of the proposal. The experiments prove that words contained in domain names do have a relation with the kind of traffic directed towards them, therefore using specifically trained word embeddings we are able to classify them in customizable categories. We also show that training word embeddings on larger natural language corpuses leads improvements in terms of precision up to 180%.

(Table 2) Estimated basement temperatures at DSDP Hole 69-504B

The achievement of deep penetration (562 m) of seafloor basalts at Hole 504B, near the Costa Rica Rift (1°13.63'N, 83°43.81'W), on DSDP Legs 69 and 70 presented a rare opportunity to examine the structure of young (6 m.y.) oceanic crust. In addition to the recovery of samples for laboratory studies, an extensive suite of downhole logs and experiments was carried out at this site, for two main purposes: (1) to allow reliable deductions about the nature of the entire section of penetrated crust, because recovery of samples was far from complete (-25%); (2) to probe the physical state of rock around the drilled hole on a scale of tens of meters to kilometers. Information on the latter large-scale phenomena at Hole 504B were provided mainly by the oblique seismic experiment, utilizing a bore-hole seismometer (Stephen 1983), and by the large-scale-electrical- resistivity experiment described below.

(Table 2) Physical properties at DSDP Holes 44-391 and 76-534

Laboratory measurements on sediment samples and density well logs run at DSDP Site 534 in the Blake-Bahama Basin were used to establish an in situ velocity and density structure. Synthetic seismograms were generated for comparison to reprocessed seismic reflection data in the vicinity of the Site. Uncertainties in the relative positions of the hole and seismic reflection data, velocity corrections, and the composition of the unrecovered section were evaluated. In light of the errors and compressed section, no unique correlation of the seismic reflection data to the drill hole is completely defensible either in this chapter or elsewhere. The preferred correlation resulting from this exercise is as follows, with the Site 534 report correlation shown in parentheses where different. Horizon beta', 887 m; Horizon beta, 950 m (975 m); Horizon C , 1202 m (1250 m); Horizon C, 1268 m (1340 m); Horizon D', 1342 m (1432 m); Horizon D, 1550 m (1552 m). The major differences in these correlations arise from the use of slightly different velocities and hole location relative to the seismic profiles. The Site 534 report results rely on hole placement on a basement flank, whereas in this chapter we locate it within a basement depression still within the uncertainty of the navigation. The Site 534 report also uses drilling rates, CDP velocity analyses, sonobuoy data, and previous similar drilling correlation methods used at Site 391, along with other geologic considerations in arriving at differing results. Although the correlation method used in this investigation is more objective and the hole location uncertainties better defined, in order to have confidence in any results we will require drilling in areas where reflections are either more widely spaced or where we have better vertical velocity control in the hole.

Annotated record of the detailed examination of Mn deposits from DODO Expedition stations

The cores and dredges described in this report were taken on the DODO Expedition in May 1964 until December 1964 by the Scripps Institution of Oceanography from the R/V Argo. A total of 290 cores and dredges were recovered and are available at Scripps for sampling and study.

(Table 1) Grain size results, statistical parameters and magnetic susceptibility values of disturbed sediments from ODP holes 204-1249C and 204-1245B

Soupy and mousse-like fabrics are disturbance sedimentary features that result from the dissociation of gas hydrate, a process that releases water. During the core retrieval process, soupy and mousse-like fabrics are produced in the gas hydrate-bearing sediments due to changes in pressure and temperature conditions. Therefore, the identification of soupy and mousse-like fabrics can be used as a proxy for the presence of gas hydrate in addition to other evidence, such as pore water freshening or anomalously cool temperature. We present here grain-size results, mineralogical composition and magnetic susceptibility data of soupy and mousse-like samples from the southern Hydrate Ridge (Cascadia accretionary complex) acquired during Leg 204 of the Ocean Drilling Program. In order to study the relationship between sedimentary texture and the presence of gas hydrates, we have compared these results with the main textural and compositional data available from the same area. Most of the disturbed analyzed samples from the summit and the western flank of southern Hydrate Ridge show a mean grain size coarser than the average mean grain size of the hemipelagic samples from the same area. The depositional features of the sediments are not recognised due to disturbance. However, their granulometric statistical parameters and distribution curves, and magnetic susceptibility logs indicate that they correspond to a turbidite facies. These results suggest that gas hydrates in the southern Hydrate Ridge could form preferentially in coarser grain-size layers that could act as conduits feeding gas from below the BSR. Two samples from the uppermost metres near the seafloor at the summit of the southern Hydrate Ridge show a finer mean grain-size value than the average of hemipelagic samples. They were located where the highest amount of gas hydrates was detected, suggesting that in this area the availability of methane gas was high enough to generate gas hydrates, even within low-permeability layers. The mineralogical composition of the soupy and mousse-like sediments does not show any specific characteristic with respect to the other samples from the southern Hydrate Ridge.

(Table T3) Vitrinite reflection of samples obtained from ODP Leg 180 sites

Seventy-one samples from Ocean Drilling Program Leg 180 sites were analyzed for vitrinite reflectance and organic type. The objective was to define maximum paleotemperatures across the western Woodlark Basin as a function of depth. The organic matter is of early Pliocene to Holocene age and was recovered from drilled depths of 4.5 to 851.3 meters below seafloor. Organic matter is generally restricted to woody fragments within the sediment, although in a number of fine-grained samples, organic matter is dispersed throughout the sample. Virtually all samples contain vitrinite, part of which may be derived from drifted logs. One sample was found to be barren of organic matter, and two contain only fusinite and semifusinite. Variation of vitrinite reflectance is not systematic with either depth or location, and it appears that formation temperatures have been insufficient to cause an increase in vitrinite reflectance levels. Textural variations within the vitrinite show better correlation with depth. Samples of hypautochthonous peats represent either a terrestrial phase of sedimentation or large peat intraclasts within the section, possibly produced by forest fires in the source areas of the organic matter. The vitrinite and peat-derived samples appear to come from eucalyptus forest settings away from the coastline. Liptinite is not abundant in most of the samples (excluding suberinite associated with woody tissues). Marine liptinite is rare to absent, although many of the samples contain abundant foraminiferal tests. Pyrite is abundant in many of the wood fragments, and some pyritization of woody tissues has taken place.

Milankovitch frequency peaks of ODP Site 113-693 (Table 3)

The first well logs collected below the Antarctic circle were obtained during Leg 113 at Site 693 on the Dronning Maud Land Margin (Antarctica) in the Weddell Sea. Gamma-ray, resistivity, and sonic logs were collected between 108.0 and 439.0 mbsf. The downhole logs show good agreement with the data collected from cores and provide a continuous measurement of the sedimentary record. These continuous log records show that the rather uniform Tertiary lithology seen in cores is characterized by high-frequency variability in the log data. Several thin hard streaks are identified, the largest of which coincides with a major Miocene hiatus. Associated with this hiatus is a change to lower illite content (and correspondingly lower gamma-ray counts) and to a significant increase in diatom content. Spectral analysis of the logs was performed on the lower Pliocene through upper Oligocene interval (108.0-343.0 mbsf). Between 108.0 and 245.0 mbsf, average sedimentation rates (50 and 26 m/m.y.) are high enough to show that variance is present in the orbital eccentricity (~95 k.y.) and obliquity (~41 k.y.) bands. Between 253.0 and 343.0 mbsf, the sedimentation rate (8 m/m.y.) is too low to resolve high frequency variations. The Milankovitch frequencies are best developed in the resistivity logs. Resistivity is responding to changes in porosity, which in these sediments is controlled by the abundance of biosiliceous sediments, particularly diatoms. The orbital forcing suggested by the Milankovitch frequencies may be influencing diatom productivity by inducing oscillations in upwelling, ice coverage, pack ice, and/or polynya. Although variations in diatom abundance were observed in the cores, they were not attributed to a Milankovitch signal, and therefore in this environment, downhole logs are an important contribution to the detection and understanding of orbitally influenced changes in sedimentation.

Grain size analysis on core plugs of sediment core CRP-1 (Table 1, 2)

Twenty four core samples from CRP-1, seven from Quaternary strata (20-43.55 meters below sea floor or mbsf) and seventeen from early Miocene strata (43.55 to 147.69 mbsf), have been analysed for their grain-size distribution using standard sieve and Sedigraph techniques. The results are in good agreement with estimates of texture made as part of the visual core description for the 1 :20 core logs for CRP-1 (Cape Roberts Science Team, 1998). Interpretation of the analyses presented here takes into account the likely setting of the site in Quaternary times as it is today, with CRP-1 high on the landward flank of a well-defined submarine ridge rising several hundred metres above basins on either side. In contrast, seismic geometries for strata deposited in early Miocene times indicate a generally planar sea floor dipping gently seaward. Fossils from these strata indicate shallow water depths (< 100 m), indicating the possibility that waves and tidal currents may have influenced sea floor sediments. The sediments analysed here are considered in terms of 3 textural facies: diamict, mud (silt and clay) and sand. Most of the Quaternary section but only 30% of the early Miocene section is diamict, a poorly sorted mixture of sand and mud with scattered clasts, indicating little wave or current influence on its texture. Although not definitive, diamict textures and other features suggest that the sediment originated as basal glacial debris but has been subsequently modified by minor winnowing, consistent with the field interpretation of this facies as ice-proximal and distal glaciomarine sediment. Sediments deposited directly from glacier ice appear to be lacking. Mud facies sediments, which comprise only 10% of the Quaternary section but a third of the early Miocene section, were deposited below wave base and largely from suspension, and show features (described elsewhere in this volume) indicative of the influence of both glacial and sediment gravity flow processes. Sand facies sediments have a considerable proportion of mud, normally more than 20%, but a well-sorted fine-very fine sand fraction. In the context of the early Miocene coastal setting we interpret these sediments as shoreface sands close to wave base.

Annotated record of the detailed examination of Mn deposits from the FANFARE Expedition stations

The cores described in this report were taken on the FANFARE Expedition in July 1959 by Scripps Institution of Oceanography from the R/V H. M. Smith and the R/V Spencer F. Baird. A total of 49 cores and dredges were recovered and are available at Scripps for sampling and study. The coring sites, all in the eastern tropical central Pacific.

Ice rafted debris distribution in sediment core CRP-1

Lonestone abundances in CRP-1 were investigated using three methods: core examination at Cape Roberts Camp, analysis of digital core images and follow-up core examination. For all images of split-core, we determined size and depth of every detectable lonestone larger than 3 mm. Lonestone abundance decreases exponentially with clast size. Although no significant depth-dependent variations in lonestone size distribution were detected, a strong 0.5-0.7 m abundance periodicity, of unknown origin, is evident within diamicts. Lonestone volume percentage was estimated from size distribution: most size classes contribute approximately the same volume to the total. Sizes >16 mm have rare enough lonestones that their counts are nonrepresentative when based on short intervals of split core. This problem does not affect total counts significantly, but the volume analysis needs to be confined to <= 6 mm lonestones to avoid instability induced by rare and nonrepresentative larger lonestones. If lonestone abundance can be used as an indicator of glacial proximity, then our CRP-1 lonestone abundance logs confirm the overall character of previously inferred variations in relative distance to the ice margin. Large-scale changes in lonestone abundance also reflect the CRP-1 sequence stratigraphy, with individual sequences generally characterised by basal lonestone-rich diamict overlain by lonestone-poor sands and muds. The relationship between glacial proximity and lonestone abundance within diamicts and within sand-mud intervals is, however, less certain. For example, two or three gradual lonestone increases may indicate regressions during glacial advances, in contrast to the more common CRP-l pattern of dominantly transgressive sequences.

Geochemistry of ODP Hole 128-798B

The capability of determining elemental concentrations through geochemical logging has recently been established. However, the quality of these data obtained in some environments has yet to be quantified. We assess the quality of geochemical logs compared with XRF results from a suite of core samples from Hole 798B. The resulting core/log correlations are only fair, because the tool has been adversely affected by the very high porosity of the formation. The results, however, do fall within the statistical uncertainties predicted by the processing. The recent application of a modified boron sleeve to the Ocean Drilling Program's geochemical logging tool is shown to reduce interference of borehole chlorine on the resultant chemistry.

Temperature measured in permafrost borehole at ICDP site 5011-3 in El'gygytgyn Crater, 2008-2011

This study focuses on the temperature field observed in boreholes drilled as part of interdisciplinary scientific campaign targeting the El'gygytgyn Crater Lake in NE Russia. Temperature data are available from two sites: the lake borehole 5011-1 located near the center of the lake reaching 400 m depth, and the land borehole 5011-3 at the rim of the lake, with a depth of 140 m. Constraints on permafrost depth and past climate changes are derived from numerical simulation of the thermal regime associated with the lake-related talik structure. The thermal properties of the subsurface needed for these simulations are based on laboratory measurements of representative cores from the quaternary sediments and the underlying impact-affected rock, complemented by further information from geophysical logs and data from published literature. The temperature observations in the lake borehole 5011-1 are dominated by thermal perturbations related to the drilling process, and thus only give reliable values for the lowermost value in the borehole. Undisturbed temperature data recorded over more than two years are available in the 140 m deep land-based borehole 5011-3. The analysis of these observations allows determination of not only the recent mean annual ground surface temperature, but also the ground surface temperature history, though with large uncertainties. Although the depth of this borehole is by far too insufficient for a complete reconstruction of past temperatures back to the Last Glacial Maximum, it still affects the thermal regime, and thus permafrost depth. This effect is constrained by numerical modeling: assuming that the lake borehole observations are hardly influenced by the past changes in surface air temperature, an estimate of steady-state conditions is possible, leading to a meaningful value of 14 ± 5 K for the post-glacial warming. The strong curvature of the temperature data in shallower depths around 60 m can be explained by a comparatively large amplitude of the Little Ice Age (up to 4 K), with low temperatures prevailing far into the 20th century. Other mechanisms, like varying porosity, may also have an influence on the temperature profile, however, our modeling studies imply a major contribution from recent climate changes.

Heat flow determinations of ODP Hole 111-504B

Hole 504B in the eastern equatorial Pacific has been the focus of five scientific drilling expeditions since it was first drilled in 1979. During these five legs, a series of temperature logs has been obtained over a time span of almost 8 yr, documenting the geothermal and hydrologic state of the oceanic crust in this region. Immediately following reentry at the onset of ODP Leg 111 operations, a high-resolution temperature probe was lowered into the borehole and a precise record of temperature vs. depth in Hole 504B was recorded down to 1300 mbsf. As was observed during previous legs, the temperature gradient in the upper 400 m was reduced, indicating that downhole flow of cool ocean waters through the casing continued, though at a diminished rate. As subhydrostatic pressures in the upper basement have gradually diminished, the volume of flow has decayed from an estimated 6000-7000 L/hr in late 1979 to about 80 L/hr during Leg 111. At depths below 480 mbsf, a predominantly conductive heat transfer environment enabled the temperature gradient log to be analyzed with respect to lithology on both fine and broad scales. Anomalies in the gradient log in the cased section through the sedimentary column were found to correspond to biostratigraphic age markers and/or sharp changes in sediment composition and texture. Broad variations in temperature gradient within the basement correlated with large-scale porosity trends. Conductive heatflow estimates depict a systematic reduction with depth, ranging from approximately 196 mW/m**2 in the sediments to 120 ± 17 mW/m**2 at 1300 mbsf. Possible causes for this observation were examined from several perspectives, but none was suitably convincing. A fluid instability analysis indicated the likely existence of convection cells within the borehole and substantiated the hypothesis of mixing within the borehole postulated from isotopic and chemical studies of borehole waters. However, such mixing of borehole fluids does not provide an adequate explanation for the heatflow variations, and the disparity between surficial and deep values of heat flow remains unresolved.

(Table T2) Remanent anomaly and Koenigsberger ratio of ODP holes

Magnetic field strength and magnetic susceptibility were logged with the geological high-resolution magnetic tool (GHMT) at three of the holes drilled during Ocean Drilling Program Leg 178 to the west of the Antarctic Peninsula. Polarity stratigraphies derived from the GHMT logs bear close resemblance to the polarities determined from core paleomagnetism at two of the holes and were used for magnetostratigraphic dating, especially in intervals where no core was recovered. Polarity is determined in the following way. First, the susceptibility log is used to determine the induced magnetization of the sediment. Then the background field, the field of the metal drill pipe, and the field anomaly of the sediment's induced magnetization are removed from the measured total field to leave the downhole anomaly of the sediment's remanent magnetization. The sign (positive or negative) of this anomaly gave a good polarity stratigraphy for Holes 1095B and 1096C, which are located in sediment drifts. A further step, correlation analysis, is based on the fact that in an interval of normal polarity sediment the remanent anomaly will correlate with the induced anomaly, whereas in reversed polarity sediment they will anticorrelate. The magnetite-rich, fine-grained sediments found in the two holes drilled into the sediment drift have a ratio of remanent to induced magnetization (the Koenigsberger ratio) of ~1. In contrast, the coarser-grained diamict sediments on the shelf have a Koenigsberger ratio of ~0.2, and extracting the remanent part of the downhole anomaly is much more difficult. By the comparison of core and log results, we can assess the viability of the GHMT polarities in detail, what proportion of the overprint in the cores is imparted by the coring process, and whether any paleointensity information is extractable from the GHMT logs.