<p>Measured data from our speed tests.</p>

This paper is distributed with data from our speed measurements in csv-table format, packed in a zip file (S1_dataset.zip), as supplementary information. The supplementary data go beyond what has been shown in the figures, with details and further explorative measurements. The general formatting paradigm of the csv tables is as follows: the column separator is a single tabulator sign; the first row contains column headings; the first column contains run ids or experiment descriptors (where average values and sample standard deviations, where calculated, are denoted as “Average” and “Stddev”, respectively); strings are not quoted and do not contain spaces; and all cells but those of the first column and row contain values in MB/s; the decimal sign is a point. the column separator is a single tabulator sign; the first row contains column headings; the first column contains run ids or experiment descriptors (where average values and sample standard deviations, where calculated, are denoted as “Average” and “Stddev”, respectively); strings are not quoted and do not contain spaces; and all cells but those of the first column and row contain values in MB/s; the decimal sign is a point. The data in the eight files contained in S1_dataset.zip is as follows: (A) Dataset_hayek25_TestA_cf_fig23_icp_LRZ_to_IT4I.csv Contains the speed-test results corresponding to the “LRZ to IT4I (with icp)” part of Fig 4 (note: these data/plot values are repeated in Fig 5). The data columns contain measured values for a file size of 5, 100 and 1 000 MB, respectively. Before the file provides the mere aggregated data for the figures (average, sample standard deviation) in the last two rows, it provides the data from all 20 runs (as executed to obtain average and sample standard deviation) in the first 20 data rows. (B) Dataset_hayek25_TestB_cf_fig2_icp_IT4I_to_LRZ.csv Is structurally equal to file A), but contains the data for the opposite transfer direction (IT4I to LRZ), used in Fig 4. (C) Dataset_hayek25_TestC_cf_fig3_B2SAFE_LRZ_to_IT4I.csv Contains the speed-test results corresponding to the “LRZ to IT4I (B2SAFE)” part of Fig 5. While the file is structurally similar to file A), note that it contains one more measurement row for a file size of 10 000 MB to verify the maximum speed and speed saturation behavior as described in the main text. (D) Dataset_hayek25_TestD_B2SAFE_IT4I_to_LRZ.csv Contains speed-test results corresponding to those in file C), but in the opposite direction (IT4I to LRZ with B2SAFE – not shown in the figure) and only extending to file sizes of 1 000 MB as in the tests A) and B). (E) Dataset_hayek25_TestE_cf_fig3_PyPut_LRZ_to_IT4I.csv Contains the speed-test results corresponding to the “LRZ to IT4I (Python Cl. ...)” part of Fig 5. The file is equal to file C) in structure. (F) Dataset_hayek25_TestF_PyGet_IT4I_from_LRZ.csv Contains speed-test results corresponding to those in file E), but in the opposite direction (IT4I to LRZ – not shown in figure). Note that these results, for practical reasons, were obtained by executing the get routine on LRZ side (i.e. we pulled data from IT4I side instead of pushing as in all other tests). Thus, the results are methodically not strictly comparable to those of file E). (G) Dataset_hayek25_TestG_cf_fig4_PyModPut_LRZ_to_IT4I.csv Contains all speed-test results corresponding to Fig 6, based on our modified version of the put routine of the Python iRODS Client. The data columns contain measured values for a file size of 10, 100, 1 000 and 10 000 MB, respectively. We have included no single-run values here, as the rows are reflecting the buffer-size dependency (average and sample standard deviation for a buffer size of 1, 10 and 100 MB). (H) Dataset_hayek25_TestH_PyModPut_IT4I_to_LRZ.csv Contains speed-test results corresponding to those in file G), but in the opposite direction (IT4I to LRZ – not shown in figure). The file is equal to file G) in structure. (A) Dataset_hayek25_TestA_cf_fig23_icp_LRZ_to_IT4I.csv Contains the speed-test results corresponding to the “LRZ to IT4I (with icp)” part of Fig 4 (note: these data/plot values are repeated in Fig 5). The data columns contain measured values for a file size of 5, 100 and 1 000 MB, respectively. Before the file provides the mere aggregated data for the figures (average, sample standard deviation) in the last two rows, it provides the data from all 20 runs (as executed to obtain average and sample standard deviation) in the first 20 data rows. (B) Dataset_hayek25_TestB_cf_fig2_icp_IT4I_to_LRZ.csv Is structurally equal to file A), but contains the data for the opposite transfer direction (IT4I to LRZ), used in Fig 4. (C) Dataset_hayek25_TestC_cf_fig3_B2SAFE_LRZ_to_IT4I.csv Contains the speed-test results corresponding to the “LRZ to IT4I (B2SAFE)” part of Fig 5. While the file is structurally similar to file A), note that it contains one more measurement row for a file size of 10 000 MB to verify the maximum speed and speed saturation behavior as described in the main text. (D) Dataset_hayek25_TestD_B2SAFE_IT4I_to_LRZ.csv Contains speed-test results corresponding to those in file C), but in the opposite direction (IT4I to LRZ with B2SAFE – not shown in the figure) and only extending to file sizes of 1 000 MB as in the tests A) and B). (E) Dataset_hayek25_TestE_cf_fig3_PyPut_LRZ_to_IT4I.csv Contains the speed-test results corresponding to the “LRZ to IT4I (Python Cl. ...)” part of Fig 5. The file is equal to file C) in structure. (F) Dataset_hayek25_TestF_PyGet_IT4I_from_LRZ.csv Contains speed-test results corresponding to those in file E), but in the opposite direction (IT4I to LRZ – not shown in figure). Note that these results, for practical reasons, were obtained by executing the get routine on LRZ side (i.e. we pulled data from IT4I side instead of pushing as in all other tests). Thus, the results are methodically not strictly comparable to those of file E). (G) Dataset_hayek25_TestG_cf_fig4_PyModPut_LRZ_to_IT4I.csv Contains all speed-test results corresponding to Fig 6, based on our modified version of the put routine of the Python iRODS Client. The data columns contain measured values for a file size of 10, 100, 1 000 and 10 000 MB, respectively. We have included no single-run values here, as the rows are reflecting the buffer-size dependency (average and sample standard deviation for a buffer size of 1, 10 and 100 MB). (H) Dataset_hayek25_TestH_PyModPut_IT4I_to_LRZ.csv Contains speed-test results corresponding to those in file G), but in the opposite direction (IT4I to LRZ – not shown in figure). The file is equal to file G) in structure. (ZIP)