>
> Thanks for your suggestion. I did not follow this:
>
> "For this use-case I would suggest using single cache puts (the same way
> you
> insert data to Oracle) and combine it with write-behind store writing to
> HDFS, this should give you better latencies."
>
> Are you suggesting not using IGFS and using Ignite cache directly with a
> write-behind store?
> May I use a regular file system as a write behind with Ignite cache?
>
Yes, if you do not need such an abstraction as files and directories, using
cache directly instead of IGFS will be faster. As for the file system, you
should not use local file system as a persistence media because
IgniteCacheStore is supposed to be a centralized storage.
>
> >If you still want to proceed with IGFS, I would try increasing IGFS file
> >block size, see FileSystemConfiguration#blockSize.
>
> What is the default block size of IGFS? I could not find it on Javadoc.
>
> Also, I am persisting in IGFS as follows:
>
> * IgniteFileSystem fs;
> IgfsPath path;
> byte[] data;
> //...
>
> try (OutputStream out = fs.create(path, true)) { out.write(data); }
> *
> If I used buffered I/O could I see significant drop in latency?
>
The default block size in IGFS is 64kb, so increasing it to 128kb or 256kb
can probably improve the latency. Since you are writing the whole data
chunk in one call, I do not think buffered output stream will make any
difference here.
Hope this helps,
AG
