Capacity of Interference-limited Three Dimensional CSMA Networks

Transcription

1 Capacity of Iterferece-limited Three Dimesioal CSMA Networks Tao Yag, Guoqiag Mao School of Electrical ad Iformatio Egieerig The Uiversity of Sydey {tao.yag, Wei Zhag School of Electrical Egieerig ad Telecommuicatios The Uiversity of New South Wales Abstract I this paper, we study the throughput of iterferece-limited three dimesioal (D) CSMA etworks. Specifically, we cosider a etwork with a total of odes uiformly i.i.d. i a cube of edge legth. Further, CSMA radom access scheme is employed ad the SINR model is used to simulate a successful trasmissio. We first give a sufficiet coditio o the trasmit power required for the CSMA etwork to be asymptotically almost surely (a.a.s.) coected as uder the SINR model. The, we demostrate costructively that a throughput of Θ is obtaiable by each ode for ( log ) a arbitrarily chose destiatio. I. INTRODUCTION Wireless multi-hop etworks have bee icreasigly used i military ad civilia applicatios. I may a applicatios, the regio i which the etwork is deployed is better modeled by a D space, istead of a two dimesioal (D) plaar area. Examples iclude a wireless etwork deployed across differet floors iside a buildig coectig a variety of devices such as computers, smart phoes, sesors etc, a etwork formed by Umaed Aerial Vehicles ad groud devices for recoaissace ad surveillace, ad uderwater acoustic sesor etworks. Capacity of such etworks is a importat problem. The scalig behavior of capacity whe the etwork becomes sufficietly large is of particular iterest. Existig work o the capacity of wireless multi-hop etworks has maily focused o the aalysis of D etworks [], [], icludig [] which cosidered D CSMA etworks. Limited work has cosidered the properties of D etworks where cetralized/determiistic schedulig schemes like TDMA are employed [], [4]. O the other had, CSMA schemes, which make use of distributed/radomized medium access protocols, has become prevailig with widespread adoptio. With CSMA, each ode checks the status of the wireless chael before sedig a packet. If the chael is idle (i.e. o carrier is detected withi its carrier-sesig rage), the the ode starts its trasmissio, otherwise, defers it, usually by a radom amout of time, util the chael becomes idle agai. Potetial trasmitters i the viciity of a active trasmitter are kept This work is partially fuded by ARC Discovery project: DP This material is based o research partially sposored by the Air Force Research Laboratory, uder agreemet umber FA The U.S. Govermet is authorized to reproduce ad distribute reprits for Govermetal purposes otwithstadig ay copyright otatio thereo. The views ad coclusios cotaied herei are those of the authors ad should ot be iterpreted as ecessarily represetig the official policies or edorsemets, either expressed or implied, of the Air Force Research Laboratory or the U.S. Govermet. off. Wireless sigals trasmitted at the same time mutually iterfere with each other. The SINR (sigal to iterferece plus oise ratio) model has bee widely used to capture the impact of iterferece o the quality of a lik ad a trasmissio is cosidered to be successful iff a miimum SINR requiremet has bee met []. Therefore, it is atural to expect CSMA could improve the etwork performace by alleviatig the iterferece. I this paper, we cosider a CSMA etwork with odes uiformly i.i.d. i a cube of edge legth ad ivestigate the throughput of the etwork. The cotributios of this paper are: ) We derive a upper boud o the iterferece experieced by ay receiver i the D CSMA etwork. Usig the result, we show that for a arbitrary SINR requiremet, there exists a trasmissio rage R 0 such that ay two odes are directly coected if their Euclidea distace is less tha or equal to R 0. Based o that, we give a sufficiet coditio o the trasmit power for the CSMA etwork to be a.a.s. coected uder the SINR model as. A coected etwork is a prerequisite for the etwork to achieve a o-zero throughput. ) We further show that i the D CSMA etwork, a throughput of Θ is achievable. Compared ( log ) with the results i [] ad [4], which showed that a throughput of Θ is attaiable by usig ( log ) either a determiistic schedulig [], [4] or without cosiderig the SINR requiremet for successful trasmissios [4], our result shows that a throughput of Θ is also attaiable eve whe CSMA is ( log ) used ad a miimum SINR is required. The remaider of this paper is orgaized as follows: Sectio II reviews related work; Sectio III defies the etwork ad metrics beig ivestigated; i Sectio IV, we give a sufficiet coditio o the trasmit power to have a a.a.s. coected D CSMA etwork; I Sectio V, we obtai a lower boud o the achievable throughput of D CSMA etworks; Sectio VI cocludes the paper ad discusses future work. II. RELATED WORK Existig work o studyig capacity problem focused maily o D etworks. The semial work [] showed that i a etwork with a total of odes distributed o a disk of

2 uit area uder the SINR model, the throughput obtaiable by each ode is ) Θ log if odes are radomly i.i.d. ad destiatio is radomly chose for each ode; ) Θ if odes locatio, traffic patter ad trasmissio rage are optimally arraged. Sice this pioeerig work, extesive efforts have bee made to ivestigate the capacity i differet scearios. Sigificat outcomes have bee achieved for both static etworks [5] [7] ad mobile etworks [8], [9]. The paper [9] showed that mobility of odes ca be exploited to sigificatly improve etwork capacity at the expese of delay. Other work i the area icludes [0], [] studied the capacity of etworks with ifrastructure support, ad [] showed that radomly placed base statios ca also boost the throughput. For etworks usig distributed/radom CSMA scheme, the recet work [] showed that a throughput of Θ ca be achieved i D CSMA etworks with radomly chose destiatios. The result is i the same order as the TDMA etwork cosidered i []. [], [] studied the iteractios betwee the trasmit power, the carrier-sesig rage ad the capacity i D CSMA etworks. Very limited work (see [], [4] ad refereces therei) has studied the capacity of D etworks ad all focused o etworks employig determiistic scheduligs. The paper [] cosidered a etworkdeployed i a sphere ad showed that a throughput of Θ is feasible. A more ( log ) recet work [4] studied the capacity of D etworks uder two scearios, i.e. odes are regularly placed ad odes are Poissoly distributed. III. NETWORK MODELS AND PRELIMINARIES We cosider a etwork with odes uiformly i.i.d. i a cube with edge legth. A. Iterferece model Assume all odes use a commo trasmit power P. Let x k,k Γ, be the locatio of ode k, where Γ represets the set of idices of all odes i the etwork. A trasmissio from ode i to ode j is successful iff the SINR at ode j is above a threshold β, i.e. SINR (x i, x j )= P(x i, x j ) N 0 + k T i P(x k, x j ) β () where T i Γ deotes the subset of odes trasmittig at the same time as ode i. (x i, x j ) represets power atteuatio from x i to x j ad assumes a power-law form, i.e., (x i, x j )=x i x j () where is the Euclidea orm ad α is the path-loss expoet. We assume that the backgroud oise N 0 is egligibly small, i.e. N 0 = 0. This assumptio is justified because iterferece is a major factor that weakes performace i wireless etworks, while the backgroud oise is typically small ad ca be combated by icreasig the trasmit power. As commoly doe i the capacity aalysis [] [4], [9], [], the impact of small-scale fadig is igored. Sice CSMA typically require a ACK packet to ackowledge a successful trasmissio, we explicitly cosider bidirectioal lik oly i the etwork. I other words, a trasmissio from ode i to ode j is successful iff both SINR (x i, x j ) ad SINR (x j, x i ) are above β. I that case, we also say that ode i ad j are directly coected. B. Defiitio of throughput The chael rate of a trasmissio from ode i to ode j is related to the associated SINR by Shao theorem, i.e., R (x i, x j )=B log ( + SINR (x i, x j )) () where B is the badwidth of the chael i Hertz. Due to the miimum SINR requiremet i (), the chael rate betwee a pair of directly coected odes is at least B log ( + β). Every ode seds data at a rate (bits/sec) to a radomly chose destiatio. A ode is both a source ad a destiatio ode for aother ode. Therefore the total umber of source-destiatio pairs is. The per-ode throughput, deoted by λ (), is defied as the maximum rate that could be achieved by ay source-destiatio pair simultaeously. A throughput of λ () is feasible if there is a temporal ad spatial schedulig scheme such that every ode ca sed λ () bits/sec o average to its destiatio, i.e. there exists a sufficietly large positive umber τ such that i every fiite time iterval [(j ) τ,jτ] every ode ca sed τλ() bits to its destiatio. A throughput is of order Θ(f ()) bits/sec if there are determiistic costats 0 < c < c < + such that lim Pr (λ () =cf () is feasible) = ad lim Pr (λ () =c f () is feasible) <. C. CSMA radom access scheme I CSMA etworks, two odes, say i ad j, are allowed to trasmit simultaeously if they ca ot detect each other s trasmissio, i.e. both P(x i, x j ) ad P(x j, x i ) are uder a certai detectio threshold P th (this is also termed as the pairwise carrier-sesig decisio model i []). This mechaism imposes a miimum separatio costrait amog the cocurret trasmitters, kow as the carrier-sesig rage. It readily follows from () that the carrier-sesig rage R c is give by R c =(P/P th ) /α (4) Uder the carrier-sesig costrait, multiple odes coted for a opportuity to trasmit ad at a particular time istat, there ca oly be oe ode trasmittig i a geographic regio determied by R c. Therefore, the chael rate give by () is shared by several odes i the viciity over time. Next, we describe how to obtai the time-average chael rate (or equivaletly the log-term chael rate i []) for each ode. Same as that i [], we cosider a idealized CSMA scheme. Assume that each ode maitais a coutdow timer, which is iitialized to a o-egative radom iteger. The timer of a ode couts dow whe the ode seses the chael idle, otherwise it is froze. A ode iitiates its trasmissio whe its timer reaches zero ad the chael is idle. After fiishig trasmissio, the ode resets its timer to a ew

3 radom iteger. The average coutdow time ca be distict for differet odes, which ca be set to cotrol the state trasitio probabilities i the ext paragraph. The above CSMA scheme ca be modeled by a Markov chai with state space S, where a state S S represets the active trasmitter set at a particular time istat. A trasitio betwee two distict states S, S S ca possibly occur iff S = {i} S for i Γ. Trasitio S {i} S (where i/ S) represets the evet that ode i will starts its trasmissio after its timer couts dow to zero. Trasitio {i} S S represets the evet that ode i fiishes its trasmissio ad hece become silet agai. Let υ be the set of state trasitio probabilities ad deote the above Markov chai by S,υ. The the time-average chael rate available for each ode ca be characterized by the statioary distributio of S,υ [4] [, Lemma 8]. IV. CONNECTIVITY OF D CSMA NETWORKS For ay throughput to be feasible, a prerequisite is that there exists a path betwee each pair of source ad destiatio, i.e. the etwork is coected. I this sectio, we first derive a upper boud o the iterferece experieced by ay receiver i the etwork. We further show that for a arbitrarily chose β, there exists a trasmissio rage R 0 such that a pair of odes are directly coected if their Euclidea distace is smaller tha or equal to R 0. Based o that, we give a sufficiet coditio o the trasmit power for the D CSMA etwork to be a.a.s. coected as uder the SINR model. The followig lemma gives a upper boud o the iterferece. Lemma. Cosider a CSMA etwork with odes arbitrarily distributed i a regio i where the carrier-sesig rage is R c, give by (4). Deote by r 0 the Euclidea distace betwee a arbitrary receiver ad its iteded trasmitter ad r 0 <R c. Whe the path loss expoet α>, the maximum iterferece is upper bouded by N (r 0 ), where N (r 0 ) = P (R c r 0 ) + 7P R c r 0 + 9P 5 α 79 α+89 Rc 7 R c r 0 (α )+54r0 α (5) (α ) (α ) (α ) Rc R c r 0 Proof: See Appedix. Remark. Note that the upper boud i Lemma applies to arbitrary ode distributio i. Further, the requiremet that α> is for the iterferece to be bouded by a costat idepedet of. If α, the the iterferece give by () approaches ifiity as. I that case, a upper boud o iterferece ca still be foud by usig the techique preseted i the proof of Lemma but that boud will be a fuctio of. I this paper, we focus o the situatio that α> to avoid some verbose but straightforward discussio o special cases that occur whe α. Corollary is a cosequece of Lemma. Corollary. Uder the same settig as that i Lemma, there exists a trasmissio rage R 0 <R c such that a pair of odes are directly coected if their Euclidea distace is smaller tha or equal to R 0, which is give implicitly by PR 0 /N (R 0 )=β () Proof: Notig that Pr 0 N(r as r 0) 0 0, Pr 0 N(r 0) Pr 0 N(r 0) 0 as r 0 Rc ad that is a mootoically decreasig fuctio of r 0, therefore there is a uique solutio to (). The rest of the proof is trivial ad hece omitted. Sice P = P th Rc α, R 0 i () ca also be expressed as a fuctio of R c. Lettig Rc R 0 β = (x ) + 7 = x, () ca be rewritte as x α 79 α+89 x 7 (α ) x+54 (α ) (α ) (α ) x x α (7) It follows that R 0 = Rc b ad b is the solutio to (7), which depeds o β ad α oly. Equatio (7) gives a more coveiet way to study the relatio betwee P ad R 0. Based o Corollary ad the result i [5, Theorem ] o the coectivity of D etworks uder the uit disk coectio model, we obtai the followig theorem. Theorem 4. Cosider a CSMA etwork with a total of odes uiformly i.i.d. i a cube with edge legth. Uder SINR model, the etwork is a.a.s. coected as if trasmit power P = P th (b ) α (log + c ()) α (8) where lim c () =+, b = b (/4π) ad >b> is the solutio to (7). Proof: The theorem readily follows from the result i [5, Theorem ] with proper scalig ad Corollary. A implicatio of Theorem 4 is that whe P is set as that i (8), a.a.s. there exists a temporal ad spatial schedulig scheme that allow ay pair of odes i the CSMA etwork to exchage packets. V. FEASIBLE THROUGHPUT I this sectio, we first describiga routig algorithm ad the show that a throughput of Θ ca be ( log ) achieved usig the routig algorithm i D CSMA etworks. Partitio the cube ito o-overlappig cubelets of edge legth s = (4 log ). Let X i be the radom umber of odes i a cubelet i. Let X = max X i ad X =mi X i where i i represets the set of idices of all cubelets. We obtai: Lemma 5. As, Pr X c log = ad Pr (X c log ) = where c = 4 + ad c = 4. Proof: Note that X i has a biomial distributio with parameters ad s ad E [X i ] = s = 4 log. Usig the Cheroff boud, we have that for ay δ (0, ), Pr [X i ( + δ) E [X i ]] exp holds. δ E[X i]

4 4 Figure. A illustratio o the umber of routes served by oe cubelet. Let δ = X, the Pr i 4 + log. There are a total of s cubelets (here we igored some trivial discussio o graularity problem caused by s ot beig a iteger). By the uio boud, we have /s lim Pr i= X i 4 + log =0. Usig a similar method, we have that for ay δ (0, ) Pr [X ( δ ) E [X]] exp holds. Takig δ E[X] δ = ad usig the uio boud yields X. A. The maximum traffic served by each cubelet For a give β, usig (7), (4), we ca choose a trasmit power so that the trasmissio rage R 0, give by Corollary, is s. This value allows ay two odes i two eighborig cubelets to directly commuicate with each other uder the SINR model. Hece, the chael rate betwee two odes, whose Euclidea distace is less tha or equal to R 0, is at least B log ( + β). Usig (7) we ca write: R c = b s, where b is the solutio to (7) for the give β. We employ a similar routig scheme to that used i [4]. For a pair of source ad destiatio odes located at (x s,y s,z s ) ad (x d,y d,z d ) respectively, packets geerated by the source are first relayed to a ode closest to (x s,y s,z d ), the to a ode closest to (x d,y s,z d ) ad fially delivered to (x d,y d,z d ). As illustrated i Fig., the shaded space represets a cubelet. Accordig to the above routig scheme, oly odes located i the three rectagular cuboid, which are bouded by dashed lies i the figure, possibly eed odes i the shaded cubelet to relay their data. Therefore the maximum umber of routes served by each cubelet is N routes = s X = X 4 log Suppose each ode seds data at a rate λ () bits/sec to its destiatio. The maximum amout of traffic each cubelet eeds to trasmit is N routes λ () at most. B. Time-average chael rate for each ode I this subsectio, we first costruct a determiistic TDMA schedulig (S t ) m t=, where S t S is the active trasmitter set durig time slot t, ad determie the time-average chael rate i [(S t ) m t= ] for ode i Γ uder this scheme. The usig the result i [, Lemma 9], we show that i [(S t ) m t= ] C ra i [S,υ] where C ra i [S,υ] is the time-average chael rate for ode i uder the CSMA scheme described i Sectio (9) III-C (modeled by the Markov chai S,υ). Fially we establish a lower boud o C ra i [S,υ]. As show i (), ay pair of directly coected odes ca trasmit at a rate at least B log ( + β). For coveiece, i the followig discussio, we cosider that the rate equals to B log ( + β) ad ormalize it to. We divide time ito slots of uit legth. It follows that the chael rate available for a particular ode i uder the schedulig scheme (S t ) m t= is equal to the fractio of time that ode i gets to trasmit, i.e. i [(S t ) m t= ]= m (i S t ) m t= We group adjacet cubelets ito o-overlappig cubes ad each cube cotais (k + ) cubelets, where k = b so that ks R c = b s. Usig Lemma 5, a.a.s. there are at most c log odes i every cubelet. Based o the above discussio, a determiistic schedulig algorithm ca be desiged such that withi time slots from t =to t =(k + ) c log, each ode gets at least oe time slot to trasmit while the set of cocurret trasmitters meets the CSMA costraits. Deote by S t the cocurret trasmitter set durig time slot t. It follows that S t S for t fractio of time spet o each S t,t, (k + ) c log ad the is, (k + ) c log. Lettig m =(k + (k+) c log ) c log, it the follows that there is a determiistic schedulig that ca achieve a timeaverage chael rate of at least B log(+β) (k+) c log i [(S t ) m t= ] B log ( + β) (k + ) c log for ode i, i.e. (0) Usig the result i [, Lemma 9] which states that there exists a properly desiged CSMA scheme that delivers suitable state trasitio probabilities v, such that for each ode i, the followig holds C ra i [S,υ] i [(S t ) m t= ] () where i [(S t ) m t= ] i () is the time-average chael rate available for ode i uder a determiistic schedulig scheme. Combiig (), (0) ad (), it ca be established that for ode i Γ, C ra i [S,υ] B log (+β). (k+) c log Lemma 5 also tells that the miimum umber of odes i every cubelet is greater tha or equal to c log a.a.s. Therefore, the miimum time-average chael rate available for each cubelet uder CSMA scheme is (c log ) B log ( + β) (k + ) c log C. Lower boud o throughput = c B log ( + β) c (k + ) () For ay per-ode throughput λ () to be feasible, the traffic load for each cubelet should ot exceed the time-average chael rate available for each cubelet, i.e., N routes λ () c B log ( + β) c (k + ) bits/sec which results i a lower boud o the feasible per-ode throughput. This is summarized i the Theorem, which forms aother major cotributio of this paper.

5 5 Figure. Desest sphere packig. Theorem. For the cosidered CSMA etworks, there exists a determiistic costat c > 0, idepedet of α, P th ad β, such that a per-ode throughput λ () = bits/sec is feasible a.a.s. as, c log (+β) (k+) ( log ) where k = b ad b is the solutio to (7) for a give β. VI. CONCLUSION I this paper, we studied the throughput of D CSMA etworks. We first provided a sufficiet coditio o the trasmit power for havig a a.a.s. coected D CSMA etwork uder the SINR model. The, usig a simple routig scheme, we obtaied a per-ode throughput of Θ ( log ) is feasible eve whe distributed/radom access scheme is used ad a miimum SINR for each successful trasmissio is specified. It remais our future work to study the optimum access scheme that maximizes the per-ode throughput. APPENDIX: PROOF OF LEMMA The derivatio of the upper boud o iterferece is similar to that used i []. The differece is i that, here we use desest sphere packig i D space to derive the upper boud. Costruct a coordiate system such that the origi o is at a trasmitter w. Cosider a trasmissio from w to its receiver u located at u ad defie r 0 u. Draw a sphere of radius R c / cetered at each cocurret trasmitter. The two spheres cetered at two closest trasmitters caot overlap. The maximum iterferece happes whe these spheres are placed i the desest way, which is to place the sphere ceters at the vertices of a face-cetered cubic lattice [7, p.9]. See Fig. for a illustratio. Group the sphere ceters ito tiers of icreasig distace from the origi. The picture (a) i Fig. shows the spheres i the st tier (i dark shade) ad the d tier (i light shade) whose ceters are located o the x y plae. All of the spheres i the d tier whose ceters are above the x y plae (icludig those o the x y plae) are show i light shade i picture (c). Although Fig. oly shows the packig alog +z axis, the packig goes alog z axis as well i the same way as alog +z axis. The umber of iterferers i the j th tier is 7j +. Let x i be the locatio of a iterferer. The miimum distace from a iterferer i the j th tier to the origi is jr c. Deote by I (r 0 ) the iterferece at ode z. Sice x i u x i r 0, () we obtai I (r 0 ) P (R c r 0 ) α + 7P R α + c r 0 j= P 7j + jr α () c r 0 The first two items i RHS of () accout for the iterferece caused by the st tier iterferers. Let U j,j=,...,, be radom variables uiformly ad i.i.d. i j,j+. It follows from the covexity of 7j + jr c r 0 whe j,α> ad Jese s iequality that P 7j + jr c r 0 j= = P 7E (U j ) + E (U j) R c r 0 j= 7U P E j + U jr c r 0 = P j= j+/ j= j / 7x + xr c r 0 dx = 9P 5 α 79 α+89 Rc 7 R c r 0 (α )+54r0 α (4) (α ) (α ) (α ) Rc R c r 0 Substitutig (4) ito (), Lemma is proved. REFERENCES [] C.-K. Chau, M. Che, ad S. C. Liew, Capacity of large-scale csma wireless etworks, IEEE/ACM Tras. Netw., vol. 9, o., pp , 0. [] P. Gupta ad P. R. Kumar, The capacity of wireless etworks, IEEE Tras. If. Theory, vol. 4, o., pp , 000. [], Iterets i the sky: capacity of d wireless etworks, i Proc. IEEE CDC, 000. [4] P. Li, M. Pa, ad Y. Fag, The capacity of three-dimesioal wireless ad hoc etworks, i Proc. IEEE INFOCOM, 0. [5] O. Dousse, M. Fraceschetti, ad P. Thira, O the throughput scalig of wireless relay etworks, IEEE Tras. If. Theory, vol. 5, o., pp. 75 7, 00. [] M. Fraceschetti, O. Dousse, D. N. C. Tse, ad P. Thira, Closig the gap i the capacity of wireless etworks via percolatio theory, IEEE Tras. If. Theory, vol. 5, o., pp , 007. [7] A. Ozgur, O. Leveque, ad D. N. C. Tse, Hierarchical cooperatio achieves optimal capacity scalig i ad hoc etworks, IEEE Tras. If. Theory, vol. 5, o. 0, pp , 007. [8] G. Sharma, R. Mazumdar, ad B. Shroff, Delay ad capacity trade-offs i mobile ad hoc etworks: A global perspective, IEEE/ACM Tras. Netw., vol. 5, o. 5, pp , 007. [9] M. Grossglauser ad D. N. C. Tse, Mobility icreases the capacity of ad hoc wireless etworks, IEEE/ACM Tras. Netw., vol. 0, o. 4, pp , 00. [0] P. Li ad Y. Fag, Impacts of topology ad traffic patter o capacity of hybrid wireless etworks, IEEE Tras. If. Theory, vol. 8, o., pp , 009. [] J.-w. Cho, S.-L. Kim, ad S. Chog, Capacity of iterferece-limited ad hoc etworks with ifrastructure support, IEEE Commu. Letters, vol. 0, o., pp. 8, 00. [] T.-S. Kim, H. Lim, ad J. C. Hou, Uderstadig ad improvig the spatial reuse i multihop wireless etworks, IEEE Tras. Mobile Comput., vol. 7, o. 0, pp. 00, 008. [] T.-Y. Li ad J. C. Hou, Iterplay of spatial reuse ad sir-determied data rates i csma/ca-based, multi-hop, multi-rate wireless etworks, i Proc. IEEE INFOCOM, 007. [4] X. Wag ad K. Kar, Throughput modellig ad fairess issues i csma/ca based ad-hoc etworks, i Proc. IEEE INFOCOM, 005. [5] V. Ravelomaaa, Extremal properties of three-dimesioal sesor etworks with applicatios, IEEE Tras. Mobile Comput., vol., o., pp. 4 57, 004. [] T. Yag, G. Mao, ad W. Zhag, Coectivity of wireless csma multihop etworks, i Proc. IEEE ICC, 0. [7] J. H. Coway ad N. J. A. Sloae, Sphere Packigs, Lattices ad Groups, rd ed. New York: Spriger, 999.