From yueyinliang at gmail.com Mon Jan 21 19:41:26 2008 From: yueyinliang at gmail.com (Yinliang Yue) Date: Tue, 22 Jan 2008 08:41:26 +0800 Subject: [Disksim-users] Which queue for scheduling scheme ? Message-ID: In disksim, there are multiple queues, for example, the queue for device drives, the queue for controls and the queue for disks. I think the queue related to multiple disk scheduling schems (e.g. SSTP, SCAN and so on) are the queue for device drives not the queue for disks. This is also imposed in the reference manual of disksim 3.0 implicitly. However, let us have a look at RAID1, there are multiple disks in one RAID. so there are multiple disk heads, and then there are multiple positioning times for a specific request for a data, because of the mirror of data. I think there must be a selection algorithm in the driver of RAID to deciding which replica will be read. Would anyone tell me whether the conclusion i guess is right or not ? The scheduling schemes are really implemented in device drivers not in disks ? -------------- next part -------------- An HTML attachment was scrubbed... URL: From ganger at ece.cmu.edu Mon Jan 21 21:25:46 2008 From: ganger at ece.cmu.edu (Greg Ganger) Date: Mon, 21 Jan 2008 21:25:46 -0500 (EST) Subject: [Disksim-users] Which queue for scheduling scheme ? In-Reply-To: References: Message-ID: There are multiple queues, including one at each disk and one for each driver-level device at the simulated device driver level. Each has a scheduling algorithm. Good luck, Greg On Tue, 22 Jan 2008, Yinliang Yue wrote: > In disksim, there are multiple queues, for example, the queue for device > drives, the queue for controls and the queue for disks. > > I think the queue related to multiple disk scheduling schems (e.g. SSTP, > SCAN and so on) are the queue for device drives not the queue for disks. > This is also imposed in the reference manual of disksim 3.0 implicitly. > > However, let us have a look at RAID1, there are multiple disks in one RAID. > so there are multiple disk heads, and then there are multiple positioning > times for a specific request for a data, because of the mirror of data. > I think there must be a selection algorithm in the driver of RAID to > deciding which replica will be read. > > Would anyone tell me whether the conclusion i guess is right or not ? > > The scheduling schemes are really implemented in device drivers not in disks > ? > From yueyinliang at gmail.com Tue Jan 22 07:52:24 2008 From: yueyinliang at gmail.com (Yinliang Yue) Date: Tue, 22 Jan 2008 20:52:24 +0800 Subject: [Disksim-users] the event in DiskSim Message-ID: When we use external trace, the number of events are much larger than that of the items of iotrace file. Could anyone tell me what's the priciples of issuing how many events ? Cound anyone tell me whether we can control the number of events by adjusting some parameters ? Thanks! -------------- next part -------------- An HTML attachment was scrubbed... URL: From swschlosser at gmail.com Tue Jan 22 08:56:44 2008 From: swschlosser at gmail.com (Steve Schlosser) Date: Tue, 22 Jan 2008 08:56:44 -0500 Subject: [Disksim-users] the event in DiskSim In-Reply-To: References: Message-ID: <4d362c350801220556i1d920271w10370e13eebd8453@mail.gmail.com> One IO request will generate many internal DiskSim events. For example, there are separate events as the request moves up and down the controller/bus stacks, for seeks, rotational latency, and data transfer, the caches, and probably a bunch more. These are all internal and have no effect on the external results (other than the fact that they make the simulation work). One of the reasons that IOs are broken into many fine-grained events is so that new IO requests can be interleaved into the simulation at any time while other requests are outstanding. -steve On Jan 22, 2008 7:52 AM, Yinliang Yue wrote: > When we use external trace, the number of events are much larger than that > of the items of iotrace file. Could anyone tell me what's the priciples of > issuing how many events ? Cound anyone tell me whether we can control the > number of events by adjusting some parameters ? > Thanks! > _______________________________________________ > Disksim-users mailing list > Disksim-users at ece.cmu.edu > https://sos.ece.cmu.edu/mailman/listinfo/disksim-users > >