E. Harris and K. Ramamohanarao. Join algorithm costs revisited. The VLDB Journal, 5(1):64--84, January 1996.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....Later, the objects in the same bucket are joined. A variation of Hash Join, called Grace Hash Join, works in two phases. In the first phase, both datasets are mapped to buckets that reside on disk. In the second phase, one bucket is read at a time and a hash table is constructed from it (see [20] for a comparison) Partition Based Spatial Join (PBSM) 38] splits the data space into files using a grid. The files of each data set are then assigned to P partitions using a round robin or hashing scheme. The partitions corresponding to the same locations are then checked to find the candidate ....

E.P. Harris and K. Ramamohanarao. Join algorithm costs revisited. VLDB Journal, 5(1):64 84, 1996.

....Later, the objects in the same bucket are joined. A variation of Hash Join, called Grace Hash Join, works in two phases. In the first phase both datasets are mapped to buckets that reside on disk. In the second phase one bucket is read at a time and a hash table is constructed from it (see [21] for a comparison) Partition Based Spatial Join (PBSM) 38] splits the data space into tiles using a grid. The tiles of each data set are then assigned to P partitions using a round robin or hashing scheme. The partitions corresponding to the same locations are then checked to find the candidate ....

E.P. Harris and K. Ramamohanarao. Join algorithm costs revisited. VLDB Journal, 5(1):64--84, 1996.

....to augment the timestamp partitioning to incorporate dynamic buffer allocation, though it is not clear at the outset that this will yield a performance benefit, over our TP H algorithm or over ETP H. Dynamic buffer allocation for conventional joins was first proposed by Harris and Ramamohanarao [HR96] They built the cost model for nested loop and hash join algorithms with the size of buffers as one of the parameters. Then, for each algorithm, they computed the optimal or suboptimal, but still good, buffer allocation that led to the minimum join cost. Finally, the optimal buffer allocation ....

E. P. Harris and K. Ramamohanarao. Join Algorithm Costs Revisited. VLDB Journal, 5(1):6484, 1996.

....Unfommately, this work mainly focuses on disk I O for different storage models (direct, clustered and normalised models) and caching issues, providing no insights into query execution evaluation. Proposed model of cache assumes an infinite RAM size, which is not relevant to real systems. Research [12] proposes methods for evaluation of several join algorithms, including nested blocks, sort merge, GRACE hash and hybrid hash algorithms. These methods determine the cost taking into account not only the latency of disk reads, but also the latency of disk seeks and CPU delays. Although covering the ....

Harris, E.P., Ramamohanaro, K. Join Algorithm Costs Revisited. The VLDB Journal, 1996, Vol. 5, No. 1, P. 64-84.

....then explained and overflow treatment sketched. 6. 1 Dynamic buffer allocation Investigation of the cost of join algorithms has showed that algorithms should try to maximize the number of blocks read from disk for both relations instead of reading one relation in large parts and the other pagewise [HR96]. For the temporal histogram partition join, we use this strategy. In the partition join, this strategy also leads to better buffer utilization. The sizes of the input buffers of both relations vary in size from partition to partition depending on the ratio of the number of active tuples in each ....

E.P. Harris and K. Ramamohanarao. Join Algorithm Costs Revisited. The VLDB Journal, 5(1):64--84, 1996.

....Analysis For the creation of a Z region partitioning from a sorted stream S ,dim we define cost functions for processing times and intermediate temporary storage. Our analysis considers the TempTris algorithm and external sorting according to Z values [2] 4. 1 The Cost Model In accordance with [6] we use a cost model that takes random pages accesses and page transfers into account. Let t p be the (average case or worst case) positioning time and t t be the transfer time of a hard disk. We assume that the prefetching or write caching strategy of the file system reads or writes a physical ....

E. P. Harris and K. Ramamohanarao, "Join algorithm costs revisited," VLDB Journal, vol. 5, 1996.

....to be stored in any particular way. 5. 1 Dynamic bu er allocation Investigation of the cost of join algorithms has shown that algorithms should try to maximize the number of blocks read from disk for both relations instead of reading one relation in large parts and the other relation pagewise [HR96]. For the temporal histogram partition join, we use this strategy, which also leads to better bu er utilization. The sizes of the input bu ers of both relations vary in size from partition to partition depending on the ratio of active tuples in each relation in the current partition. The sizes of ....

E.P. Harris and K. Ramamohanarao. Join Algorithm Costs Revisited. The VLDB Journal, 5(1):64-84, 1996.

....that try to determine the best way to manage memory and schedule page fetches such that the number of I Os is minimized. In contrast, our work focuses at a more fundamental level to reduce the number of data items to be considered for join by using ordering information in the datasets. Recently, HR96] has emphasized the need to consider disk seek time, data transfer time and CPU time rather than merely considering the number of I Os for determining the cost of join processing. Hence accessing all the data items can create a lot of unnecessary overheads. This clearly shows that the benefits ....

E.P. Harris and K. Ramamohanarao. Join algorithm costs revisited. VLDB Journal, 5(1), January 1996.

....operations of a query engine. I O costs are modelled according to [HCLS97] i.e. the difference between random and sequential I O operations and the block transfer size are taken into account. CPU costs and corresponding parameter settings for hash based operations are taken from [PCV94] and [HR96] The cardinalities and record sizes of the base tables Customer, Order, and Lineitem (cf. Table 1) are chosen according to the TPC D benchmark at scale factor 1.0. In some experiments, we varied the size of the Order table in order to show how our concepts would scale for large database sizes. ....

E. Harris and K. Ramamohanarao. Join algorithm costs revisited. The VLDB Journal, 5(1):64--84, 1996.

....of our cost model is strongly influenced by the structure of modern query engines implementing the iterator model. This means that cost estimations are calculated on a per iterator basis. I O costs are modeled according to [HCLS97] and the CPU operation assumptions are mostly based on [PCV94] and [HR96] Our cost model contains extensions to deal with set valued attributes and our new P (PM) # M algorithm. Due to space limitations, we cannot discuss individual formulae. The cost formulae model disk I O quite precisely by means of differentiating between seek, latency, and transfer time. As a ....

Harris E, Ramamohanarao K (1996) Join algorithm costs revisited. VLDB J 5(1): 64--84

....the Order table, then Step 2 can be omitted altogether and Step 3 is carried out using the full Order table. The tradeoffs between generalized hash teams and partition nested loop joins are fairly much the same as between (grace and hybrid) hash joins and blockwise nested loop joins; see, e.g. HR96, HCLS97] If the Customer table is large and must be partitioned into many partitions, partition nested loop joins are likely to perform poorly for rereading the reduced Order table many times. On the other hand, partition nested loop joins might perform better than generalized hash teams if it ....

E. Harris and K. Ramamohanarao. Join algorithm costs revisited. The VLDB Journal, 5(1):64--84, 1996.

....tradeoff between preaggregation and query response time still exists, but MHC reduces it significantly. 5. Performance Analysis The cost functions used for our analysis take clustering, prefetching as well as CPU time and I O time into account and were derived in [Mar99] and [MZB99] based on [HR96]. For retrieving or grouping and aggregating (i.e. sorting) a relation in combination with multidimensional hierarchical restrictions we simulated response times and intermediate temporary storage. We consider several organizations of the fact table of a star schema: MHC, a composite secondary ....

E.P. Harris and K. Ramamohanarao. Join algorithm costs revisited. VLDB Journal, 5, 1996.

....of our cost model is strongly influenced by the structure of modern query engines implementing the iterator model. This means that cost estimations are calculated on a per iterator basis. I O costs are modeled according to [HCLS97] and the CPU operation assumptions are mostly based on [PCV94] and [HR96] Our cost model contains extensions to deal with set valued attributes and our new P(PM ) # M algorithm. Due to space limitations, we cannot discuss individual formulae. The cost formulae model disk I O quite precisely by means of di#erentiating between seek, latency, and transfer time. As a ....

E. Harris and K. Ramamohanarao. Join algorithm costs revisited. The VLDB Journal, 5(1):64--84, 1996.

....buffer T k sum of average seek and latency time T t time for transfer of one page T c time for hashing a tuple T j time for finding the join partner of a tuple Table 2: Additional parameters for cost model 3. 4 Cost model Our cost model for Diag Join is based on the cost models presented in [17]. Additional parameters needed for the cost model are presented in Table 2. The cost C I=O for transferring a set of jjR x jj pages from disk to memory, or vice versa, through a buffer size B x is given by C I=O = jjR x jj; B x ) jjR x jj B x Delta T k jjR x jj Delta T t (1) where ....

....= jR 1 j Delta T c (5) C Read RN = C I=O (jjR N jj; 1) 6) C Join = jRN j Delta T j (7) CW rite = C I=O (jjtmpF ilejj; 1) 8) The costs in the second phase depend on the join algorithm used. In our case we applied GRACE hash join in the second phase (for cost models of GRACE hash join see [15, 17]) hence C Phase2 = CGRACE (R 1 ; tmpF ile) 9) Symbol Definition N(a; b; oe) normal distribution n(x; oe) density function of normal distribution w lo starting position of window w hi ending position of window m lo starting position of middle hash table (w lo m lo ) m hi ending ....

[Article contains additional citation context not shown here]

E.P. Harris and K. Ramamohanarao. Join algorithm costs revisited. VLDB Journal, 5(1):64--84, 1996.

....query plans produced by commercial database systems for the TPC D queries. The plans were mostly left deep with group by operators sitting at the top of the plans (i.e. we did not consider any early aggregation alla [YL94, CS94] Grace hash and block nestedloop were the preferred join methods [HR96, HCLS97] and we also used hashing for most of our group by operations. The plans (including the memory allocation) we used for the compressed and uncompressed databases were identical; that is, we first found a good plan for a query using the uncompressed database, and then ran this plan on the ....

E. Harris and K. Ramamohanarao. Join algorithm costs revisited. The VLDB Journal, 5(1):64--84, January 1996.

....database still more than twice as much as P(PM) M (in spite of the faster host for the O RDBMS) 5 Analytical Evaluation We developed a cost model as vehicle for a broader analysis. I o costs are modelled according to [HCLS97] and the CPU operation assumptions are mostly based on [PCV94] and [HR96] Our cost model contains extensions to deal with set valued attributes and our new P(PM) M algorithm. Due to space limitations, we cannot discuss individual formulas. The cost formulas model disk i o quite precisely by means of differentiating between seek, latency, and transfer time. As a ....

E. P. Harris and K. Ramamohanarao. Join algorithm costs revisited. The VLDB Journal, 5(1):64--84, 1996.

No context found.

E. Harris and K. Ramamohanarao. Join algorithm costs revisited. The VLDB Journal, 5(1):64--84, January 1996.

No context found.

E.P. Harris and K. Ramamohanarao. Join algorithm costs revisited. VLDB Journal, 5(1):64--84, 1996.

No context found.

E.P. Harris, and K. Ramamohanarao. Join algorithm costs revisited. VLDB Journal, 5, 1996

No context found.

E. Harris and K. Ramamohanarao. Join algorithm costs revisited. The VLDB Journal, 5(1):64--84, January 1996.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC