Is there any way to tune the write buffer size? Or know where that exists? From: John Plevyak <[email protected]> Reply-To: "[email protected]" <[email protected]> Date: Sunday, April 28, 2019 at 8:14 PM To: "[email protected]" <[email protected]> Subject: [EXTERNAL] Re: Default TS behavior when disk write speed are too slow
Correct. ATS allows limits the outstanding async writes and then when the write buffer fills it stops enquing the data and the objects are not saved to disk. This is independent of the number of clients. On Sun, Apr 28, 2019 at 3:04 PM Jason Yang <[email protected]<mailto:[email protected]>> wrote: Thank you, John! Just to confirm my understanding is correct, ATS will proxy the requests without writing this request to disk when disk write speed cannot catch up, is that right? I am running a simulation/emulation using ATS, for the warming up period, if my fake clients send requests too fast, it is possible that the objects might be not saved to disks, right? And this is also true even if I have only one client because the write process is asynchronous, right? Jason On Apr 28, 2019, 17:53 -0400, John Plevyak <[email protected]<mailto:[email protected]>>, wrote: ATS does not delay a request if it can't write down the data. Instead it will simply proxy the request without writing it down if the write buffer becomes full because of a slow disk. On Sun, Apr 28, 2019 at 1:35 PM Jason Yang <[email protected]<mailto:[email protected]>> wrote: Hi Community, What’s the default behavior of ATS if the disk write speed is too slow, will it slow down each request? Can ATS still serve requests without storing to disk? I am thinking about the following scenario, the disk we are using occasionally is really slow and the miss ratio can be super high (meaning a large number of objects need to be written to a slow disk). Thank you! Jason
