Re: File system very slow

[vadud3@xxxxxxxxx]

I ran the smallfile utility like below for create and read and LOOKUP number looks too high

# smallfile_cli.py --top /data/shared/tmp/ --threads 8 --files 2000 --file-size 15 --record-size 15 --fsync N --operation create

# smallfile_cli.py --top /data/shared/tmp/ --threads 8 --files 2000 --file-size 15 --record-size 15 --fsync N --operation read

total threads = 8
total files = 16000
total IOPS = 125
total data = ""   0.229 GiB
100.00% of requested files processed, warning threshold is  70.00
elapsed time =   127.516
files/sec = 125.931253
IOPS = 125.931253
MiB/sec = 1.844696

# gluster volume profile atlassian info cumulative  | egrep 'Brick|LOOKUP'
Brick: node1:/data/foo/gluster
     37.43    1424.67 us      13.78 us 2244438.29 us        2024515      LOOKUP
Brick: note2:/data/foo/gluster
     53.01     413.87 us      14.03 us  262160.52 us        3122542      LOOKUP

On Tue, May 26, 2020 at 2:59 PM <vadud3@xxxxxxxxx> wrote:

I made the following changes for small file performance as suggested by http://blog.gluster.org/gluster-tiering-and-small-file-performance/

I am still seeing du -sh /data/shared taking 39 minutes.

Any other tuning I can do. Most of my files are 15K. Here is sample of small files with size and number of occurrences

FileSize.    # of occurrence

====        ============

1.1K 1122
1.1M 1040
1.2K 1281
1.2M 1357
1.3K 1149
1.3M 1098
1.4K 1119
1.5K 1189
1.6K 1036
1.7K 1169
11K 2157
12K 2398
13K 2402
14K 2406
15K 2426
16K 2386
17K 1986
18K 2037
19K 1829
2.0K 1027
2.1K 1048
2.4K 1013
20K 1585
21K 1713
22K 1590
23K 1371
24K 1428
25K 1444
26K 1391
27K 1217
28K 1485
29K 1282
30K 1303
31K 1275
32K 1296
33K 1058
36K 1023
37K 1107
39K 1092
41K 1034
42K 1187
46K 1030

On Mon, May 25, 2020 at 5:30 PM <vadud3@xxxxxxxxx> wrote:

time du -sh /data/shared

431G    /data/shared

real    45m49.992s
user    0m20.043s
sys    2m32.456s

gluster fs is extremely slow

Any suggestions on what settings to change to improve it?

--

Asif Iqbal
PGP Key: 0xE62693C5 KeyServer: pgp.mit.edu
A: Because it messes up the order in which people normally read text.
Q: Why is top-posting such a bad thing?

--

Asif Iqbal
PGP Key: 0xE62693C5 KeyServer: pgp.mit.edu
A: Because it messes up the order in which people normally read text.
Q: Why is top-posting such a bad thing?

--

Asif Iqbal
PGP Key: 0xE62693C5 KeyServer: pgp.mit.edu
A: Because it messes up the order in which people normally read text.
Q: Why is top-posting such a bad thing?

________



Community Meeting Calendar:

Schedule -
Every 2nd and 4th Tuesday at 14:30 IST / 09:00 UTC
Bridge: https://bluejeans.com/441850968

Gluster-users mailing list
Gluster-users@xxxxxxxxxxx
https://lists.gluster.org/mailman/listinfo/gluster-users