Requesting kernel parameters

[EricvanLoon]

Hi guys,

Our TSM 7.1 servers (with directory containers) were suffering from a very slow response for a long time. We already applied the kernel parameters specified in the installation guide and the SP Blueprints, but since we recently changed several other kernel parameters in one go, thing improved a lot. It’s still not great, but at least it’s not taking ages anymore to set up a new client session.
I’m interested in the value you are using on your Linux servers, especially for servers with a large workload on directory containers. If this is the case, can you please send me the output of the following commands:

sysctl vm.dirty_background_ratio
sysctl vm.dirty_ratio
sysctl vm.dirty_expire_centisecs
sysctl vm.dirty_writeback_centisecs
sysctl net.core.rmem_max
sysctl net.core.wmem_max
sysctl net.ipv4.tcp_rmem
sysctl net.ipv4.tcp_wmem
sysctl net.core.netdev_max_backlog
sysctl net.ipv4.tcp_moderate_rcvbuf
sysctl net.ipv4.tcp_no_metrics_save

Thank you VERY much for your help in advance!

Kind regards,
Eric van Loon
Air France/KLM Storage & Backup

 

[moon-buddy]

On what slowness are you exactly having issues with? Backup speed? Client-to-server communication?

Please clarify.

 

[EricvanLoon]

Very slow response on several commands like Q STGPOOL which took at least 45 seconds. Each new client or admin session also took very long to initiate, sometimes up to a minute.
We had a case open at IBM which went all up to development, but they had no clue. The only thing they saw was that the delay was in the DB2 identification part, but nor TSM support nor DB support was able to fix it.
Only after we changed all these parameters in one go, the performance improved a lot. That's why I would like to compare them with other users.
Kind regards,
Eric

 

[moon-buddy]

When you said you followed the Blueprint, I am assuming you had all bases covered.

But...

What Intel platform are you running TSM on?
# of CPU?
Local or SAN disk for the DB2 volumes and LOGS?
FC speed?
Memory size?

I am running TSM on Linux on Cisco blades and never needed to apply the tuning parameters you listed. A typical TSM server for me has:

48 CPU cores
64 or 128 GB of RAM as the case maybe
internal disk of RAID 1, 300 GB capacity for OS/TSM binaries
DB Volume/LOGS on Netapp via FC, 8 Gbps

 

[EricvanLoon]

With all due respect, I don't want to go into details here. Like stated earlier, the performance issue seems to be fixed after changing several kernel parameters in one go.
I'm just curious about the values others have set.

 

[moon-buddy]

With all due respect, I don't want to go into details here. Like stated earlier, the performance issue seems to be fixed after changing several kernel parameters in one go.
I'm just curious about the values others have set.

Like I said, I don't have to do all that tweaking.