Very low performance issue

[rFlex]

Hi,
I still have a huge performance issue with Open-E DSS v6.

My server configuration :
Xeon L5520 (4 quad cores)
8DDR3 ECC
5x Crucial RealSSD 256Gio (290Mio/s in SATA2 per SSD)
3x Intel e1000 Gigabits Ethernet

The 3 networks cards are in bonding (802.3ad) and the SSDs are in software Raid5 (using DSS) so I should get a 3Gbit/s connection and a very high bandwidth capacity for the disks (I except more than 800Mio/s in local).
But the problem is that I don't even managed to have more than 100Mio/s per iSCSI volume ! The speed is not stable and various between 20Mio/s and 100Mio/s even if the client is 2 or 3 links bonded.

I tried to connect a client to the SAN directly (without using a switch) and I still have the problems.

Thanks for guiding me

[nsc]

Hi,

For me, you can't aggregate bandwith with bonding.
You have to use MPIO.

[Al-S]

Also you did not tell us what is the build of DSSv6 that you are running now?

[rFlex]

I did the test :
*-----------------------------------------------------------------------------*
hdparm -t /dev/sda
*-----------------------------------------------------------------------------*


/dev/sda:
Timing buffered disk reads: 774 MB in 3.00 seconds = 257.96 MB/sec

(Same results on sdb, sdbc, sdd)

*-----------------------------------------------------------------------------*
hdparm -t /dev/sde
*-----------------------------------------------------------------------------*


/dev/sde:
Timing buffered disk reads: 400 MB in 3.00 seconds = 133.25 MB/sec

This is the last SSD, it is a bit slow !

*-----------------------------------------------------------------------------*
hdparm -t /dev/sdf
*-----------------------------------------------------------------------------*


/dev/sdf:
Timing buffered disk reads: 274 MB in 3.02 seconds = 90.81 MB/sec

This is the 2x 2.5" RAID1 HDD used for DSS installation

*-----------------------------------------------------------------------------*
hdparm -t /dev/md0
*-----------------------------------------------------------------------------*


/dev/md0:
Timing buffered disk reads: 1884 MB in 3.00 seconds = 627.64 MB/sec

It's a bit slow for a Raid5 of 5 SSD, maybe the problem is caused by sde...

*-----------------------------------------------------------------------------*
ethtool bond0
*-----------------------------------------------------------------------------*

Settings for bond0:
No data available

It that normal?

All eth0,eth1,eth2 are set to 1000Mbit/s and no errors occured.


Version of DSS : 6.0up75.8401.5377 64bit

[Al-S]

You need a small update for your NIC card, as you are having build up75, so please open a support ticket, and ask me to send you a Small Update files.
0852-DSS_V6_up75-Intel_1Gbps_NIC_driver_set
0853-DSS_V6_up75-Intel_10Gbps_NIC_driver_set

______________________
Alaa Souqi

[rFlex]

I'm still using the Lite version since I don't know if DSS will works well with my configuration, so I can't open a support ticket.
What should I do ? Pay for support for obtaining a patch who will maybe not solves my problems?
Thanks however

[Pi-L]

search for individual cards with ethtool, not for bond, like ethtool eth0 etc.

[Pi-L]

Since there could be few things causing lowered performance except hardware failure so the best thing to do first is to download logs from DSS (in Web GUI -> Status -> Hardware -> Logs) and search tests.log file for:
1) hdparm - to check actual raid performance
2) ethtool - to check actual network cards speed
3) ifconfig - to see if there is any packet loss (errors in network transmission)
Please take a look at it cause it may be for example cards are running only 100MBps mode for some reason.