How to improve the AWA-Filetransfer: speed up / increase data throuput

[Thomas Richert]

Hi Community,

We try to speed up the AWA-Filetransfers in our Environment. Therefore we tested various settings and found that the data throughput for a 2.5GB zip file is only 10 MB/s. Is there a possibility to increase the data throuput?
For comparison, with scp we have a throuput of 110MB/s.

Does anyone have any general information about ways to improve an AWA-Filetransfer? Unfortunately we coudn`t find anything in the AWA-Help


Thanks
Thomas

[Peter Grundler]

Hi Thomas,

it might be related to the settings in HOSTCHAR. Kindly please provide the configuration.

Thanks,
Peter

[Peter Daransky]

Hi Thomas, 
it depends a on the Agent version, but basically there is bunch of options which allows you to tune the Filetransfer. 
It's always the choice between safety and speed, so maybe not all of them are suitable for you. It's just list and you have to consider which one can be applicable for you :-) 

You have to check following setting in UC_HOSTCHAR_DEFAULT:

• FT_ASYNC_QUIT_* - with this setting you can control the real throughput, mean you can control the behavior between sending waiting for the data packets and confirmations. Similar as TCP/IP do.
• FT_USE_MD5 - causes to compute the checksum on both sides for each data, so that if the FT fails you can resume from last known position ... 

and the INI settings: 

SENDBUFFERSIZE and RECVBUFFERSIZE - where you can control how the TCP buffers are allocated.

Hope this helps. 

BR p

[Thomas Richert]

Thanks for your fast and useful help guys....

Until now we only used the synchronous transfer.
Now we have tried different settings with strong and weak asynchronous transfer.
After these tests we want to use the asynchronous file transfer in the future to get it faster.

@Peter Daransky But how does the buffersize play into the whole thing? The Help says that the buffersize must be increased accordingly. But with what values? Do you have any suggestions or guidelines?
Currently we have not set any value, so the default value applies...


​

[Peter Daransky]

Hi Thomas, 

regarding your question:  But how does the buffersize play into the whole thing?

Wish there is simple answer to this :-) The buffersize is mapped to the socket buffer size. Means it's OS and Automic Version depend. Means if you are on 12.x Automic then as much I can remember was the value for Windows 256KB. To rise the value more do not bring much benefits especially when there are many parallel FTs. For v21 do not have the values, but there is another logic (TLS) included where I'm not sure if the buffer size will change it much ? You have to try :-). 

One hint for you, the File Transfer frames are limited to max. 32KB  (now) ... At the moment the default is 1MB, this is nice when you run one FT in time ... But this looks different if you starts several transfers. 

All of the previous values was related to Windows, if you're running on Unix it will be for sure different :-(

So long story short, "sometimes less is more".

BR p