Odd SLP behavior

[Douglas_A]

After seting up an HP D2D for a customer i was runnign some tests and found an odd situation.. When writing backups it writes around 200mb/sec (agg throughput) this is fine, but if an SLP kicks off while it is still writing the throughput drops down to around 3mb/sec and stays there till the SLP is completed or i kill the job.

Whats off as well is on the Media server (windows server 2k8) the Network utilization before the SLP is roughly 80% when the SLP kicks off the utilization drops to 4-5%.

Environment is Netbackup 7.0.1 windows 2k8 master, win 2k8 media.

Reviewing the logs shows nothing at all unusual with the app, but maybe someone has somewhere i can go to find out ANYthing that might be going on?

Thanks in advance.

[Nicolai]

Sound like disk congestion to me. On my disk storage unit , performance declien if disk read and writes occur concurrent. I can't however explain the very steep figures you see. What type of arry are you using ?

[AAlmroth]

As with all disk systems, most experience degraded performance when having concurrent access, and even worse when both ways; read and write. On top of this, if you use a de-dup enabled disk system, you may experience the de-hydration overhead on read as seen on most de-duplication devices.

You may want to verify the settings on your NIC ports. I've seen dramatically improved performance by disabling chimney offloading, TCP checksums RX/TX offloading etc on Windows 2008. It seems most NIC device/drivers are not capable of providing sustained high throughput, and especially not bi-direction.

Unless you already have separate NIC port for the I/O to the disk storage, you should perhaps consider using multiple NIC ports on the media servers. Also if you use iSCSI for the D2D system, it is wise to separate the traffic.

/A