Integer Test 1: Software Installation of IRIX 6.5 (Default Configuration)

SGI Performance Comparisons

Software Installation of IRIX 6.5 (Default Configuration)

Last Change: 08/May/2008

These results came out of my study of installing IRIX 6.5, which you should read if you want to know exactly how the data was obtained. Table 1 includes timing details of all intermediate stages of the installation process. The installation test times shown on the OS installation test pages do not include the 'human' element of the installation procedure, eg. switching CDs, typing in commands, etc. When using my stopwatch to time the installations, I halted the stopwatch whenever a prompt appeared and restarted the stopwatch once the required input was entered and the procedure again underway.

You may need to widen your browser window to view the complete table.

                                   ******* Sub-Task Completion Times *******
                                   (16%) (51%)  (86%)  (91%)  (94%)   (100%)
                       CD   Read   Pre-                       Inst    End of  End of
                       ROM  Inst   inst  F 1    Apps   F 2    Tools    Exit   rqsall
CPU Type / L2               Tools  Ends  Ends   Ends   Ends   Ends     Coms   ELF libs

O2 R5000SC/200 1MB:    32   0:42   0:54  06:31  15:07  16:46  17:45   22:52   0:34:06
O2 R5000SC/200 1MB:    12   0:46   0:54  06:58  16:04  18:05  19:20   24:08   0:35:28
I2 R4400SC/250 2MB:    32   0:47   1:11  06:45  15:17  16:57  17:55   23:12   0:36:05
I2 R4400SC/250 2MB:     2   1:16   1:11  09:18  21:06  23:28  24:38   29:57   0:42:31
Indy R4400SC/200 1MB:  32   0:42   1:26  08:32  19:42  21:40  22:50   29:41   0:45:24
Indy R4400SC/200 1MB:   2   1:16   1:26  11:11  25:49  28:29  29:58   36:53   0:52:35
Indy R5000SC/180 512K: 32   0:49   1:00  09:04  20:40  22:37  23:49   31:11   0:46:55
Indy R5000PC/150:       2   1:19   1:17  12:06  27:47  32:05    ?     41:58      ?    [1]
Indy R4600PC/133:      32   0:44   1:45  11:48  28:26  31:28  33:07   46:11   1:08:45
Indy R4600PC/133:       2   1:16   1:45  14:10  33:37  37:12  39:00   52:05   1:14:39
Indy R4600PC/100:      32   0:35   2:04  12:36  30:18  33:36  35:20   48:28   1:12:11
Indy R4600PC/100:       2   1:16   2:04  15:01  35:33  39:32  41:28   54:54   1:18:36

         Table 1: Detailed Timings for a Default IRIX 6.5 OS Installation

[1] Test run by Euan Harris of Glasgow University.

Rationale

Some parts of the software installation process do not involve the CDROM or source device at all. This includes:

preinstallation check,
dependencies check,
disk usage check,
running the exit commands,
rqsall ELF libs and creating a new kernel.

It is likely that these steps involve integer accesses on disk files that are similar to the operations which occur when compiling programs. Thus, the time taken to complete these steps can be compared between systems, ie. the differences between systems for two different tests (2X vs. 32X CDROM) should be roughly the same (there will be slight differences due to the different disks being used on occasion).

Those aspects of the installation which do use the CDROM can be analysed to study how CDROM performance varies with CPU strength, and whether or not a fast CDROM is worth having given a particular CPU type (these results are discussed later).

Table 2 shows the time taken for each individual step in the installation process, ie. the times obtained by subtracting one column from the next.

                   CD   PRE-                       INST-  EXIT   RQSALL
                   ROM  INST   F1    APPS    F2    TOOLS  COMS    ELF

O2 R5000SC/200:    32   0:54  05:37  08:36  01:39  00:59  05:07  11:14
O2 R5000SC/200:    12   0:54  06:04  09:06  02:01  01:15  04:48  11:20
I2 R4400SC/250:    32   1:11  05:34  08:32  01:40  00:58  05:17  12:53
I2 R4400SC/250:     2   1:11  08:07  11:48  02:22  01:10  05:19  12:34
Indy R4400SC/200:  32   1:26  07:06  11:10  01:58  01:10  06:51  15:43
Indy R4400SC/200:   2   1:26  09:45  14:38  02:40  01:29  06:55  15:42
Indy R5000SC/180:  32   1:00  08:04  11:36  01:57  01:12  07:22  15:44
Indy R5000SC/180:   2
Indy R4600PC/133:  32   1:45  10:03  16:38  03:02  01:39  13:04  22:34
Indy R4600PC/133:   2   1:45  12:25  19:27  03:35  01:48  13:05  22:34
Indy R4600PC/100:  32   2:04  10:32  17:42  03:18  01:44  13:08  23:43
Indy R4600PC/100:   2   2:04  12:57  20:32  03:59  01:56  13:26  23:42

   Table 2: Individual times for each installation step (IRIX 6.5)

Here are the percentage differences for the paired results, ie. highest divided by lowest (these figures should be low in order to be usable):

                   PRE-  EXIT  RQSALL
                   INST  COMS   ELF

O2 R5000SC/200:     0%   6.6%   0.9%
I2 R4400SC/250:     0%   1.0%   2.5%
Indy R4400SC/200:   0%   1.0%   0.0%
Indy R4600PC/133:   0%   0.0%   0.0%
Indy R4600PC/100:   0%   2.0%   0.0%

   Table 3: Paired result differences.

These are satisfactory, indeed they're perfect in many cases. However, I may run the O2 tests again to make sure I didn't make any mistakes during the timings.

Now for the all-important comparison: the average performance for each system compared to the average performance of the slowest system (Indy R4600PC/100) for each stage (divide the target time by the slowest system's time). Thus, the numbers denote how much faster each stage was executed compared to the slowest system (ie. smaller = better).

                    PRE-   EXIT  RQSALL   AVERAGE
                    INST   COMS   ELF

O2 R5000SC/200:     0.44   0.37   0.48  |  0.44
I2 R4400SC/250:     0.57   0.40   0.54  |  0.49
Indy R4400SC/200:   0.69   0.52   0.66  |  0.62
Indy R4600PC/133:   0.85   0.98   0.95  |  0.96
Indy R4600PC/100:   1.00   1.00   1.00  |  1.00

Table 4: Comparison to Indy R4600PC/100
   performance (factor differences)

The variances between tests are probably due to the different degrees to which the disk is being accessed, the benefit of having any L2 cache to aid the computations, differences in disk speeds and latencies, different amounts of data being processed, etc.

It's very noticable that not having any L2 cache hinders the processing of the EXIT COMS and RQSALL ELF stages.

Note: to calculate percentage improvements, subtract each number from 1 and multiply by 100, eg. the RQSALL step for Indy R4600SC/200 is 34% faster than Indy R4600PC/100, while (overall) Indigo2 R4400SC/250 is 51% faster than Indy R4600PC/100 at performing these particular integer/disk tasks.