1

I have a 1Gb/s bandwidth between my computers. The source can actually read and send data at 25MB/s via SSH. My test file is 4GB, I use scp without compression for the test transfer. The rate is 25MB/s at the beginning, but after a point it starts to decrease quite suddenly. After sending 2GB the rate is down at 4.5MB/s almost steadily, the load is up to 6. Disks are WDC WD5001ABYS-01YNA0 models. Filesystem is ext3.

The bottleneck seems to be the RAID bus controller: 3ware Inc 9550SX SATA-II RAID PCI-X The tw_cli commands run very slowly (I see the output appearing line by line, does not occur when not stressed).

# tw_cli /c1 show

Unit  UnitType  Status         %RCmpl  %V/I/M  Stripe  Size(GB)  Cache  AVrfy
------------------------------------------------------------------------------
u0    RAID-5    OK             -       -       256K    931.303   OFF    OFF

Port   Status           Unit   Size        Blocks        Serial
---------------------------------------------------------------
p0     OK               u0     465.76 GB   976773168     WD-WCAS87320631
p1     OK               -      465.76 GB   976773168     WD-WCAS87223554
p2     OK               u0     465.76 GB   976773168     WD-WCAS87159042
p3     OK               u0     465.76 GB   976773168     WD-WMAYP6812676
p4     NOT-PRESENT      -      -           -             -
p5     NOT-PRESENT      -      -           -             -
p6     NOT-PRESENT      -      -           -             -
p7     NOT-PRESENT      -      -           -             -

Head of tw_cli /c1 diag

### CLI Version:      x86 (32 bit)
### Time Stamp:       00:13.27 11-Dec-2013
### Host Name:        somelinux
### OS Version:       Linux 2.6.18
### Driver Version:   2.26.02.010
### Controller ID:    1
### Model:            9550SXU-8LP
### Firmware:         FE9X 3.04.00.005
### BIOS:             BE9X 3.04.00.002
### Serial #:         L320810A7430290
### Available Memory: 112MB

As far as I know the cache cannot count in this case because the write is a long linear one. These drives can do 40MB/s writing when they are alone.

What can be the cause of all this? What should I do?

Additional info

Reading is 32MB/s with the caused load of 1.5.

Notinlist
  • 217
  • 2
  • 10

1 Answers1

3

The output from tw_cli shows that the write cache is disabled. By default, 3ware cards disable the write cache if the BBU is not present.

When the write cache is disabled each write operation must be flushed to all disks to prevent possible corruption in the event of a power failure.

Write cache can be enabled by running:

tw_cli /c1/u0 set cache=on

Beware that if you enable write cache without having a working/charged BBU or a UPS, you may experience data loss if a unexpected power failure occurs.

mpolden
  • 46
  • 2
  • How serious can the data loss be? Like some bad sectors? Or will I loose my whole array? – Notinlist Dec 13 '13 at 10:35
  • The data loss should be containted to the files being written. If filesystem metadata is being written, your filesystem might end up in a inconsistent state. This may require a fsck or similar during the next boot. – mpolden Dec 13 '13 at 18:21
  • Translate that to "full disappearing data unless you pay for recovery". Things CAN get nasty. Not likely, but possible. – TomTom Feb 19 '14 at 15:26