fstime.c - Seperate r/w files for each parallel

[pdeng6]

Existing workload is using 1 read file and 1 write file for file read/write/copy test. In multi-parallel scenario, it leads to high file lock contention, while read/write/copy is not stressed.

This change seperates r/w files for each parallel to satisfy the multi-parallel test purpose.

[pdeng6]

Test on my 160 core machine:

Without the change, 160P score is even lower than 1P.

	1P	160P	160P/1P
File Copy 1024 bufsize 2000 maxblocks	4,151	2,477	59.68%
File Copy 256 bufsize 500 maxblocks	2,816	1,554	55.16%
File Copy 4096 bufsize 8000 maxblocks	7,746	5,265	67.96%

With the change, 160P score is ~26X - 79X of 1P:

	1P	160P	160P/1P
File Copy 1024 bufsize 2000 maxblocks	4,311	249,846	57.95
File Copy 256 bufsize 500 maxblocks	2,817	222,938	79.14
File Copy 4096 bufsize 8000 maxblocks	8,224	213,497	25.96

[gstrauss]

PATH_MAX is a bit more portable on unix than is MAX_PATH (hi Windoze), but we do not need it. See below.

Let's reduce the number of lines changed in the patch:

-#define FNAME0  "dummy0"
-#define FNAME1  "dummy1"
+char FNAME0[] = "dummy0-XXXXXXXXXX";
+char FNAME1[] = "dummy1-XXXXXXXXXX";

later:

snprintf(FNAME0+sizeof("dummy0"), sizeof(FNAME0)-sizeof("dummy0"), "%d", pid);
snprintf(FNAME1+sizeof("dummy1"), sizeof(FNAME1)-sizeof("dummy1"), "%d", pid);

Note: there may still be some filesystem contention on the directory file since these are all created in the same directory.

---

meta: benchmarking, like statistics, can be manipulated. Comparing numbers generated from different code can be less accurate and even misleading. I made similar comments many years ago here:
#23 (comment)

That said, where the implementation is obviously detrimental to what the code is trying to exercise and test, I am happy to evaluate proposed patches.

[pdeng6]

Thanks @gstrauss , quite agree with the comments.
For the concern of remained contention, according to my profiling data on both ext4 and xfs, with this change, it is not obvious anymore, maybe we can address it in future if detected anywhere.

Cheers
Pan

[gstrauss]

Thank you for your contribution.