[metrics-dev] Fwd: [usage-stats] Daily Globus Usage-Stats Update

[Lee Liming]

Answers below.

         -- Lee

On Oct 12, 2006, at 2:27 PM, Thompson, Kevin L. wrote:

> Do we know:
>
> * the size of transfers
yes.

> * the elapsed time of transfers
indirectly.  We know the size of transfers and the bandwidth, so we  
could compute the elapsed time as size/bandwidth. We don't have a  
report generator that reports specifically on this.  Can you explain  
why it would be useful to know elapsed time?  Or is it actually the  
bandwidth data that you are interested in?

> * the number of striped transfers
yes, plus we know the number of stripes used.  We also know the same  
for parallel streams.

> * any context of the transfers in relation to other workflow elements
no. a general solution would require communication between software  
components that doesn't exist and isn't likely to exist anytime in  
the forseeable future.  We do know how many GridFTP transfers are  
orchestrated by RFT.  We have a plan to identify the portion of  
GridFTP requests that are specifically linked to WS GRAM requests,  
but this will require changes to the WS GRAM usage reporting because  
it doesn't currently report how many files were staged in/out.  Other  
workflow systems (e.g., Condor, VDS, etc.) do not use our usage  
reporting module, so it's impossible to know that they are behind the  
requests.

>
> kevin
>
>
> -----Original Message-----
> From: owner-metrics-dev at globus.org [mailto:owner-metrics- 
> dev at globus.org]
> On Behalf Of Lee Liming
> Sent: Thursday, October 12, 2006 3:13 PM
> To: metrics-dev at globus.org; Bill Allcock
> Subject: Re: [metrics-dev] Fwd: [usage-stats] Daily Globus Usage-Stats
> Update
>
> Bill:
>
> Here are some summary charts relevant to your points. The first shows
> the number of transfers per month, the second shows server hosts per
> month, with a breakdown by top level domain. Note that that the number
> of transfers seems to be somewhat related to the number of server  
> hosts
> each month, though the relationship isn't exactly proportional.  (That
> makes sense: some servers are used more than others and the usage of
> each server also varies from month to month.)
>
> The usual caveats apply.  We don't know how many UDP packets are being
> dropped between the servers and our receiver, and we don't know how  
> many
> GridFTP deployments have usage reporting turned off.
>
>         -- Lee
>