From wilfried.dron at gmail.com Fri Jun 5 06:23:12 2009 From: wilfried.dron at gmail.com (Wilfried DRON) Date: Fri, 5 Jun 2009 12:23:12 +0200 Subject: [Disksim-users] Compilation trouble with disksim 4.0 Message-ID: Hi, I'm working on SSD performance reshearching team at the PRISM, UVSQ's laboratory. I just get the last version of your famous simulator Disksim and when i try to compil it i got this messages: [sewes at localhost disksim-4.0]$ make make -C libddbg make[1]: Entering directory `/home/sewes/Bureau/disksim/disksim-4.0/libddbg' cc -g -I../libparam -c -o ddbg_assert.o ddbg_assert.c make[1]: cc : commande introuvable make[1]: *** [ddbg_assert.o] Erreur 127 make[1]: Leaving directory `/home/sewes/Bureau/disksim/disksim-4.0/libddbg' make: *** [all] Erreur 2 Could you help me? Is there a missing package in my OS? I'm using Mandriva linux 2009 Spring. Thank you Wilfried -------------- next part -------------- An HTML attachment was scrubbed... URL: From weafon at cs.nctu.edu.tw Tue Jun 16 19:21:25 2009 From: weafon at cs.nctu.edu.tw (weafon) Date: Tue, 16 Jun 2009 16:21:25 -0700 Subject: [Disksim-users] Negative transfer time in Seagate Cheetah 15k.5 disk spec Message-ID: <4A3828F5.7030101@cs.nctu.edu.tw> Hi, I have the same query about the negative transfer time. A negative value seems to lead the zero transfer time when the bus transfer time is zero. I think the time should approximate 0.005 ms According to the spec of the hard disk: http://www.seagate.com/docs/pdf/datasheet/disc/ds_cheetah_15k_5.pdf , the sustained transfer rate of SEAGATE_ST3146855FC is between 73 to 125. If we assume the sustain transfer rate is 99, or (73+125)/2, then the bulk sector transfer time would be 512/(99*1024*1024) = 4.932 * 10-6 s. Does anyone have some idea for the problem? Thanks, Weafon > > ---------------------- > *Maliha* rimuni at gmail.com > > > /Tue Mar 17 16:16:02 EDT 2009/ > > Hi > > I have downloaded the specification of Seagate Cheetah 15k.5 disk from > the > "on-line database of disk parameters" provided by the disksim team. The > disksim_disk block in file Seagate-Cheetah15k5.diskspecs specifies the > disk > transfer time as follows: > Bulk sector transfer time = -0.015000 > > Does anyone have any idea why the transfer time is negative? (The Bus and > Controller transfer times are specified as zero). > > Thanks in advance, > Maliha > > > From savulescucostin at yahoo.com Fri Jun 19 09:21:31 2009 From: savulescucostin at yahoo.com (Costin Savulescu) Date: Fri, 19 Jun 2009 06:21:31 -0700 (PDT) Subject: [Disksim-users] k-client problem Message-ID: <663467.83486.qm@web30105.mail.mud.yahoo.com> Good day everybody. I would like to ask you for some hints and pieces of advice about a project that I have for obtaining my Bachlor's Degree dipoloma. My project is called "Disk space administration in the case of the k-client problem". The k-client problem reffers to having multiple threads or processes requesting acces to the disk. So in essence this is a concurrency problem. What I would like to ask is how do I emphasize the k-client problem using DiskSim 4.0? I have also studied some disk access algorithms. Can anyone give me some hints? I suppose I will need to create a parameter file to simulate a disk in order to emphasize the k-client problem. Awaiting for your reply, thank you in advance. All the best! -------------- next part -------------- An HTML attachment was scrubbed... URL: From yueyinliang at gmail.com Fri Jun 19 09:34:09 2009 From: yueyinliang at gmail.com (Yinliang Yue) Date: Fri, 19 Jun 2009 21:34:09 +0800 Subject: [Disksim-users] k-client problem In-Reply-To: <663467.83486.qm@web30105.mail.mud.yahoo.com> References: <663467.83486.qm@web30105.mail.mud.yahoo.com> Message-ID: I think you should emphasize on the "disk space management", not the k-client problem. In disksim, you can create a trace file, in which the time slots among multiple I/O requests are small enough or approach to zero. You may notice that disksim support the following case: the arrive time of two or more I/O request can be equal to each other. So I think generate the trace you needed and then try to propose a new efficient disk space managenet approch to reduce the average response time or improve the throughput under the k-client currency situation. Best wishes! Yinliang HUST, China On 6/19/09, Costin Savulescu wrote: > > Good day everybody. I would like to ask you for some hints and pieces of > advice about a project that I have for obtaining my Bachlor's Degree > dipoloma. My project is called "Disk space administration in the case of the > k-client problem". The k-client problem reffers to having multiple threads > or processes requesting acces to the disk. So in essence this is a > concurrency problem. > What I would like to ask is how do I emphasize the k-client problem using > DiskSim 4.0? I have also studied some disk access algorithms. Can anyone > give me some hints? I suppose I will need to create a parameter file to > simulate a disk in order to emphasize the k-client problem. > Awaiting for your reply, thank you in advance. All the best! > > > _______________________________________________ > Disksim-users mailing list > Disksim-users at ece.cmu.edu > https://sos.ece.cmu.edu/mailman/listinfo/disksim-users > > -- Best regards! Yinliang Yue HUST, Wuhan, China -------------- next part -------------- An HTML attachment was scrubbed... URL: From ganger at ece.cmu.edu Fri Jun 19 11:35:22 2009 From: ganger at ece.cmu.edu (Greg Ganger) Date: Fri, 19 Jun 2009 11:35:22 -0400 (EDT) Subject: [Disksim-users] k-client problem In-Reply-To: References: <663467.83486.qm@web30105.mail.mud.yahoo.com> Message-ID: Alternately, you might be able to configure multiple generators in a disksim's synthetic workload generator component, with one per "client". You should cozy up to the user manual, given your specific target. Good luck! Greg On Fri, 19 Jun 2009, Yinliang Yue wrote: > I think you should emphasize on the "disk space management", not the > k-client problem. > > In disksim, you can create a trace file, in which the time slots among > multiple I/O requests are small enough or approach to zero. > > You may notice that disksim support the following case: the arrive time of > two or more I/O request can be equal to each other. > > So I think generate the trace you needed and then try to propose a new > efficient disk space managenet approch to reduce the average response time > or improve the throughput under the k-client currency situation. > > Best wishes! > > Yinliang > > HUST, China > > On 6/19/09, Costin Savulescu wrote: >> >> Good day everybody. I would like to ask you for some hints and pieces of >> advice about a project that I have for obtaining my Bachlor's Degree >> dipoloma. My project is called "Disk space administration in the case of the >> k-client problem". The k-client problem reffers to having multiple threads >> or processes requesting acces to the disk. So in essence this is a >> concurrency problem. >> What I would like to ask is how do I emphasize the k-client problem using >> DiskSim 4.0? I have also studied some disk access algorithms. Can anyone >> give me some hints? I suppose I will need to create a parameter file to >> simulate a disk in order to emphasize the k-client problem. >> Awaiting for your reply, thank you in advance. All the best! >> >> >> _______________________________________________ >> Disksim-users mailing list >> Disksim-users at ece.cmu.edu >> https://sos.ece.cmu.edu/mailman/listinfo/disksim-users >> >> > > > -- > Best regards! > > Yinliang Yue > HUST, Wuhan, China > From n080di at yahoo.com Sun Jun 21 18:05:06 2009 From: n080di at yahoo.com (Nobody Nobody Nobody) Date: Sun, 21 Jun 2009 15:05:06 -0700 (PDT) Subject: [Disksim-users] slave mode and outstanding events Message-ID: <107135.61402.qm@web55107.mail.re4.yahoo.com> Hello disksimmers! Suppose disksim is used in slave mode by a system simulator. Suppose a request is sent to disksim using the disksim_interface_request_arrive() function. Can this function be called again (i.e., can another request be sent in) before disksim has indicated completion of the previous request through the callback? If not, then is it necessary that requests be queued up by the system level simulator until disksim is ready for the next one? Thank you. -------------- next part -------------- An HTML attachment was scrubbed... URL: From zgy04 at 126.com Sun Jun 21 19:41:33 2009 From: zgy04 at 126.com (Steve) Date: Mon, 22 Jun 2009 07:41:33 +0800 Subject: [Disksim-users] slave mode and outstanding events References: <107135.61402.qm@web55107.mail.re4.yahoo.com> Message-ID: <200906220741319744956@126.com> >Hello disksimmers! > >Suppose disksim is used in slave mode by a system simulator. > >Suppose a request is sent to disksim using the disksim_interface_request_arrive() function. > >Can this function be called again (i.e., can another request be sent in) before disksim has indicated completion of the previous request through the callback? > yes, it can be called again. This is like the interruption call. >If not, then is it necessary that requests be queued up by the system level simulator until disksim is ready for the next one? > >Thank you. > > > > >_______________________________________________ >Disksim-users mailing list >Disksim-users at ece.cmu.edu >https://sos.ece.cmu.edu/mailman/listinfo/disksim-users > = = = = = = = = = = = = = = = = = = = = ?? From zgy04 at 126.com Sun Jun 21 20:13:39 2009 From: zgy04 at 126.com (=?utf-8?B?U3RldmU=?=) Date: Mon, 22 Jun 2009 08:13:39 +0800 Subject: [Disksim-users] =?utf-8?q?slave_mode_and_outstanding_events?= References: <774266.36195.qm@web55107.mail.re4.yahoo.com> Message-ID: <200906220813387716546@126.com> It is My pleasure! Yes. the callback routines will be called multiple times. I suggest that you should refer to two source files: the src/syssim_driver.c and src/syssim_driver.h. Tell me if any further progress. ======= 2009-06-22 07:58:32 Nobody Nobody Nobody Written: ======= > >Thank you. > >The example (by Eran G. included in the disksim/src/), however, mentions that only one callback can be outstanding. If I have multiple requests inside disksim, then disksim could call the callback scheduler multiple times. Is that OK? > > > >--- On Sun, 6/21/09, Steve wrote: > >From: Steve >Subject: Re: [Disksim-users] slave mode and outstanding events >To: "Nobody Nobody Nobody" , "disksim-users" >Date: Sunday, June 21, 2009, 7:41 PM > >>Hello disksimmers! >> >>Suppose disksim is used in slave mode by a system simulator. >> >>Suppose a request is sent to disksim using the disksim_interface_request_arrive() function. >> >>Can this function be called again (i.e., can another request be sent in) before disksim has indicated completion of the previous request through the callback? >> >yes, it can be called again. This is like the interruption call. > > >>If not, then is it necessary that requests be queued up by the system level simulator until disksim is ready for the next one? >> >>Thank you. >> >> >> >>? ? ? >>_______________________________________________ >>Disksim-users mailing list >>Disksim-users at ece.cmu.edu >>https://sos.ece.cmu.edu/mailman/listinfo/disksim-users > >> > >= = = = = = = = = = = = = = = = = = = = >??? ??? ??? > >?? > > > > = = = = = = = = = = = = = = = = = = = = ?? From n080di at yahoo.com Mon Jun 22 19:32:34 2009 From: n080di at yahoo.com (Nobody Nobody Nobody) Date: Mon, 22 Jun 2009 16:32:34 -0700 (PDT) Subject: [Disksim-users] Fw: Re: slave mode and outstanding events Message-ID: <893609.44752.qm@web55106.mail.re4.yahoo.com> Disksimmers! Another follow-up question: ? When disksim reports completion, it uses a function with the following signature: void* (SysTime t, struct disksim_request *r, void *ctx) ? What does it return (if anything useful) in *r and *ctx? Preliminary tests show that r does not appear to contain details about the request that was completed by disksim. Thank you. --- On Sun, 6/21/09, Nobody Nobody Nobody wrote: From: Nobody Nobody Nobody Subject: [Disksim-users] slave mode and outstanding events To: disksim-users at ece.cmu.edu Date: Sunday, June 21, 2009, 6:05 PM Hello disksimmers! Suppose disksim is used in slave mode by a system simulator. Suppose a request is sent to disksim using the disksim_interface_request_arrive() function. Can this function be called again (i.e., can another request be sent in) before disksim has indicated completion of the previous request through the callback? If not, then is it necessary that requests be queued up by the system level simulator until disksim is ready for the next one? Thank you. -----Inline Attachment Follows----- _______________________________________________ Disksim-users mailing list Disksim-users at ece.cmu.edu https://sos.ece.cmu.edu/mailman/listinfo/disksim-users -------------- next part -------------- An HTML attachment was scrubbed... URL: From yueyinliang at gmail.com Wed Jun 24 05:46:48 2009 From: yueyinliang at gmail.com (Yinliang Yue) Date: Wed, 24 Jun 2009 17:46:48 +0800 Subject: [Disksim-users] fpcheck error in disksim_ctlr.c Message-ID: Dear all, I encounter the following errors when I issue some requests from a trace file to disksim: *assertion failed: in disk_buffer_sector_done() (disksim_ctlr.c : **** ) : currdisk->fpcheck :* *Aborted. * Could anyone tell me how to resolve this problem ? Any help is appreciated. -- Best regards! Yinliang Yue HUST, Wuhan, China -------------- next part -------------- An HTML attachment was scrubbed... URL: From courageous at gmail.com Mon Jun 29 13:44:30 2009 From: courageous at gmail.com (Krithika) Date: Mon, 29 Jun 2009 13:44:30 -0400 Subject: [Disksim-users] splicing extra requests to simulation Message-ID: Hi, I wan't to add some extra r/w requests during a trace driven simulation. Currently what I do is, I create a ioreq_event structure and feed it to disksim_simulate_event() function. The disksim_run_simulation() function iteratively calls disksim_simulate_event(). So, on some conditions, that function would call disksim_simulate_event(event), where 'event' is my ioreq_event. Otherwise it would call disksim_simulate_event(NULL), in which case it takes the normal execution path of getting an event from trace and executing it. My problem is, after my extra requests are executed, when a normal request is starting to execute, I get an error disksim_controller.c:364: controller_bus_delay_complete: Assertion `trv->next == curr' failed I think when I change the request flow, there is some mismatch in the stats maintained by controller structure (controller->buswait). Can anyone tell me what is the best way to splice extra requests to the simulation ? Or if I already do things correctly, please let me know how to get around such kind of errors.. Thank you very much, Krithika. -------------- next part -------------- An HTML attachment was scrubbed... URL: From zgy04 at 126.com Mon Jun 29 20:42:19 2009 From: zgy04 at 126.com (Steve) Date: Tue, 30 Jun 2009 08:42:19 +0800 Subject: [Disksim-users] splicing extra requests to simulation References: Message-ID: <200906300842179575963@126.com> you can write a system level simulator with disksim as a slave module. Your problem will get easier. All the best. Steve zgy04 at 126.com 2009-06-30 ======= 2009-06-30 01:44:45 Krithika Written: ======= >Hi, > >I wan't to add some extra r/w requests during a trace driven simulation. >Currently what I do is, I create a ioreq_event structure and feed it to >disksim_simulate_event() function. The disksim_run_simulation() function >iteratively calls disksim_simulate_event(). So, on some conditions, that >function would call disksim_simulate_event(event), where 'event' is my >ioreq_event. Otherwise it would call disksim_simulate_event(NULL), in which >case it takes the normal execution path of getting an event from trace and >executing it. > >My problem is, after my extra requests are executed, when a normal request >is starting to execute, I get an error > >disksim_controller.c:364: controller_bus_delay_complete: Assertion >`trv->next == curr' failed > >I think when I change the request flow, there is some mismatch in the stats >maintained by controller structure (controller->buswait). > >Can anyone tell me what is the best way to splice extra requests to the >simulation ? Or if I already do things correctly, please let me know how to >get around such kind of errors.. > >Thank you very much, >Krithika. > >_______________________________________________ >Disksim-users mailing list >Disksim-users at ece.cmu.edu >https://sos.ece.cmu.edu/mailman/listinfo/disksim-users > = = = = = = = = = = = = = = = = = = = = From lizarazu_007 at hotmail.com Tue Jun 30 05:27:13 2009 From: lizarazu_007 at hotmail.com (lagenda pedang) Date: Tue, 30 Jun 2009 17:27:13 +0800 Subject: [Disksim-users] Tracefile cannot be opened for read access Message-ID: hi all, i'm new with disksim. i try to run the example command in the manual that is: ./disksim parms.1B stdout ascii t.Jan6 0 "disk1 .. disk16" "Segment size (in blks)" 64 "disk*" "Scheduler:Scheduling policy" 4 but it give the output like this: *** Output file name: stdout *** Input trace format: ascii Tracefile t.Jan6 cannot be opened for read access can someone help me point out where that i'm going wrong? thank you in advance. regards, lizarazu. _________________________________________________________________ NEW! Get Windows Live FREE. http://www.get.live.com/wl/all -------------- next part -------------- An HTML attachment was scrubbed... URL: