Hi all, though I've been working with FN in production for 3 years now (and I LOVE IT) I still consider myself a newbie in some aspects.
Right now I was trying to build a new FN box using this hardware:
- HP DL 180 G5 (Old 2u Server)
- 2 IBM M1015 cards (IT flashed with latest version, sas2flash output: SAS2008(B2) 20.00.07.00 14.01.00.08)
- 12 Intel DC S3610 800GB SSD drives in 6 2-drive mirrors (similar to RAID 10)
- Also tried and alternative 5x3TB WD Red with an Intel DC S3500 80GB ZIL SSD
- 16 GB ECC RAM (I know it's a low amount, the server can't handle more)
- Server has PCIe v1.0 4x and 8x
- 10 GB fibre network card
- 1 GB Ethernet network card
Also one of my production backup server is:
- IBM SystemX3200 M2
- 1 IBM M1015 cards (IT flashed with latest version, sas2flash output: SAS2008(B2) 20.00.07.00 14.01.00.08)
- 6x3TB WD Red with an Intel DC S3500 80GB ZIL SSD in RaidZ2
- 16 GB ECC RAM
- Server also has PCIe v1.0 4x and 8x
- 1 GB Ethernet network card
Situation: I created a Jail and installed Bonnie++ in both this new server and the mentioned production server
In both I run:
Results are:
Notice that the Prod Server has a 50% higher performance in every aspect than the new one, and the configuration in the HDD scenario is fairly similar. I concluded that there's an issue with the m1015 cards and the new (HP) server. I verified:
- IRQ (assignment, collisions, etc and though I 'm no expert I didn't find any thing suspicious and confs look fairly similar in both servers)
- left only one m1015 card on new server to perform the HDD test and placed it in a 8x slot. Same stuff.
- checked with lspci -vvv on both servers, found no significant (to my eyes) differences.
If someone can point me in the right direction as to how can I troubleshoot this I would highly appreciate it.
Briefing all the above info:
- I seem to be having some kind of bottleneck in the HP server, when I look at the reports in the FN UI while running the tests I see 10MB/s speeds on the Prod server and 2 to 4 MB/s speed on the new one, and I can't figure why. Same disks, same card, similar server specs...
Thanks! (and sorry if I didn't comply with anything on this post, I haven't posted much)
Matias.
Right now I was trying to build a new FN box using this hardware:
- HP DL 180 G5 (Old 2u Server)
- 2 IBM M1015 cards (IT flashed with latest version, sas2flash output: SAS2008(B2) 20.00.07.00 14.01.00.08)
- 12 Intel DC S3610 800GB SSD drives in 6 2-drive mirrors (similar to RAID 10)
- Also tried and alternative 5x3TB WD Red with an Intel DC S3500 80GB ZIL SSD
- 16 GB ECC RAM (I know it's a low amount, the server can't handle more)
- Server has PCIe v1.0 4x and 8x
- 10 GB fibre network card
- 1 GB Ethernet network card
Also one of my production backup server is:
- IBM SystemX3200 M2
- 1 IBM M1015 cards (IT flashed with latest version, sas2flash output: SAS2008(B2) 20.00.07.00 14.01.00.08)
- 6x3TB WD Red with an Intel DC S3500 80GB ZIL SSD in RaidZ2
- 16 GB ECC RAM
- Server also has PCIe v1.0 4x and 8x
- 1 GB Ethernet network card
Situation: I created a Jail and installed Bonnie++ in both this new server and the mentioned production server
In both I run:
Code:
bonnie -d /test -s 64000
Results are:
Code:
-------Sequential Output-------- ---Sequential Input-- --Random-- -Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks--- New server - SSD K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU /sec %CPU 245921 93.8 333140 90.3 257534 91.7 214043 99.8 658649 99.8 12265.8 244.2 Prod srvr - HDD K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU /sec %CPU 360194 90.4 517828 87.4 399126 90.1 321813 99.3 945104 96.7 12128.0 138.5 New server - HDD K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU /sec %CPU 241881 93.3 349314 93.8 269634 94.6 214645 99.9 663871 99.8 12350.7 250.7
Notice that the Prod Server has a 50% higher performance in every aspect than the new one, and the configuration in the HDD scenario is fairly similar. I concluded that there's an issue with the m1015 cards and the new (HP) server. I verified:
- IRQ (assignment, collisions, etc and though I 'm no expert I didn't find any thing suspicious and confs look fairly similar in both servers)
- left only one m1015 card on new server to perform the HDD test and placed it in a 8x slot. Same stuff.
- checked with lspci -vvv on both servers, found no significant (to my eyes) differences.
If someone can point me in the right direction as to how can I troubleshoot this I would highly appreciate it.
Briefing all the above info:
- I seem to be having some kind of bottleneck in the HP server, when I look at the reports in the FN UI while running the tests I see 10MB/s speeds on the Prod server and 2 to 4 MB/s speed on the new one, and I can't figure why. Same disks, same card, similar server specs...
Thanks! (and sorry if I didn't comply with anything on this post, I haven't posted much)
Matias.