CX4-120, new raid groups slow disk writes.

did you check if write cache is configured and enabled ?  How's read speed ?

its enabled on both SPs currently at 60% low and 80% high. Disk read is at around 180Mb/s. Can be this low write speeds related to the zeroing of this disks ?

i have not tested performance right after RG creation but i would not expect it to be very significant, should be a background task. How are you testing write performance ?

Im using a dd :

[root@xxxxxxxx ~]# dd if=/dev/zero of=/dev/mapper/IQ_MAIN bs=128 count=50000

50000+0 records in

50000+0 records out

6400000 bytes (6.4 MB) copied, 0.32338 s, 19.8 MB/s

dirty caching has been configured to 0 in order to flush directly to disk :

vm.dirty_background_ratio = 0

vm.dirty_background_bytes = 0

vm.dirty_ratio = 0

vm.dirty_ratio = 10

### added dirty_expire and dirty_writeback

vm.dirty_expire_centisecs=500

vm.dirty_writeback_centisecs=100

vm.dirty_bytes = 0

Not sure what else i could check maybe the LUN configuration is what is causing the issues ?

Thanks!

Write performance depends on a lot of different factors - the amount of Write Cache Memory assigned - it should be much larger than Read Cache Memory, the design of the Raid Group - Raid 5, Raid 6, Raid 10, the speed of the disks in the raid group - FC (either 10K RPM or 15K RPM) or SATA (7200 RPM, the host conection type  - FC or iSCSI, the speed of the connection, etc.

If you could provide a more detailed description of the type of disks used in the raid group, the raid type that would help.

The rating on the different disks in the CX4 are below - these are called the "rule-of-thumb" that EMC uses to help determine how a particular workload will behave. The Throughput numbers are based on a workload that is small block (less than 32KB IO Size), random IO with Read/Write ratio of 80/20. The Bandwidth is based on IO Size of 64KB or larger.

Throughput IO/sec

15K RPM FC = 180 IOPS

10K RPM FC = 120 IOPS

7200 RPM SATA = 80

Bandwidth MB/s

15K RPM FC = 12-14MB/s

10K RPM FC = 10-12MB/s

7200 RPM SATA = 8MB/s

All of these are general in nature and are affected by other activities occurring on the array. See the following document for a more detailed looked at how performance is measured.

https://www.emc.com/collateral/hardware/white-papers/h5773-clariion-best-practices-performance-availability-wp.pdf

glen

Thanks glen for the info provided - appreciate that. Already had a look at the doc and its quite informative, re-IO/s and bandwidth, i'm aware about how much speed each disk type is able to provide & that is what is strange here is that even though we have multiple SATA disks I cannot reach more than the mentioned 20MB/s of speed. Below is the RAID groups that we have together with disk types and number of disks in each :

RAID GROUP 0 - Raid 1/0 Type FC , 2 disks

RAID GROUP 1 - Raid 1/0 Type FC , 8 disks

RAID GROUP 2 - Raid 5 Type Sata2, 5 disks

RAID GROUP 3 - Raid 5 Type Sata2, 4 disks

RAID GROUP 4 - Sata2 , hotspare

RAID GROUP 5 - FC , hotspare

On top of these we created the LUNs that should be used by our platform , basically R1/0 LUN1, on RAID GROUP 0, R1/0 LUN 2 on RAID GROUP1 and then we have multiple R5 LUNs on RAID GROUP 2 , and 1 R5 LUN on RAID GROUP 3.

Thanks!

Etienne.

Let's take RG 2 as an example - this is a 4+1 R5 - each disk can handle about 6MB/s - with 5 disks that would be about 30MB/s - without anything else going on in that raid group.

You could try using IOmeter as a better test. Set it up to run 100% Writes, 100% Sequential, 64KB IO Size - this would give you the best possible test. But IOMeter uses "Workers" and the more threads you generate, the faster the performance. So set up at least 8 workers and Set the Queuing to 10. Single threaded tests are not a good measure of an array's potential as it is designed specifically to work best when using multiple threads.

glen