Hi Matevz,
Yes and no. A read request and include a pre-read list pretty much of any
size (well, up to 2MB or about 256K requests). The read itself is not
required to read any data (i.e. have 0 length so that the pre-read list is
the only meaningful part). The server will then start to fetch the
fragments into the filesystem cache. This makes sense for HD devices not
so much for SSD devices. In fact, it's not clear this makes sense at all
but it's there. The "no" part is that the proxy server does not pass this
information along. To prefetching stops at the proxy server. See,
http://www.xrootd.org/doc/prod/XRdv299.htm#_Toc337053393
for details.
Andy
On Tue, 23 Oct 2012, Matevz Tadel wrote:
> Hi,
>
> Is there a way for a client to issue an advisory prefetch request to a
> server?
> The scenario I have in mind is:
> 1. The client knows it will need to access a specific part of the file but
> it will take it a while to plow through, with many small read requests.
> 2. It suspects (or knows) it is actually talking to a proxy server that
> needs to fetch data over WAN.
> 3. It knows it will NOT need any other part of the file.
> The request would be ignored by normal (non-proxy) servers.
>
> The use case is populating proxy-cache with known missing blocks from hadoop,
> 64 or 128MB in size ... but doing it on demand, when the "broken" file is
> actually accessed, and still trying to minimize latency in retrieval.
>
> Cheers,
> Matevz
>
> ########################################################################
> Use REPLY-ALL to reply to list
>
> To unsubscribe from the XROOTD-DEV list, click the following link:
> https://listserv.slac.stanford.edu/cgi-bin/wa?SUBED1=XROOTD-DEV&A=1
>
########################################################################
Use REPLY-ALL to reply to list
To unsubscribe from the XROOTD-DEV list, click the following link:
https://listserv.slac.stanford.edu/cgi-bin/wa?SUBED1=XROOTD-DEV&A=1
|