[Disksim-users] Requests stripped at I/O driver for RAID-0

Yipkei Kwok ykwok2 at miners.utep.edu
Sun Feb 26 23:16:49 EST 2012


Hi,

On real hardware, requests are supposed to be stripped in the RAID
controller for a RAID-0 configuration. However, it is not the case in
DiskSim 4.0. When a request arrives the I/O driver, iodriver_request()
calls logorg_maprequest() to divide the request into stripes. Then,
iodriver_request() calls handle_new_request() to schedule each stripe
to the corresponding I/O-driver queue. The number of I/O driver queues
equals the number of disks and each stripe is sent to the queue
corresponding to the disk that is going to service the stripe.

1. Is it a problem with my parameter file? I copied below the
parameters of the I/O driver, controller, topology, and logical
organization below. I am attaching the entire parameter file in this
email but I am not sure if the mailing list server allows attachments.

2. If it is not because of a mistake in my parameter file, is there
any reason when DiskSim is designed this way?

Please help. Thank you.




disksim_iodriver DRIVER0 {
type = 1,
Constant access time = 0.0,
   Scheduler = disksim_ioqueue {
      Scheduling policy = 1,
      Cylinder mapping strategy = 1,
      Write initiation delay = 0.0,
      Read initiation delay = 0.0,
      Sequential stream scheme = 0,
      Maximum concat size = 0,
      Overlapping request scheme = 0,
      Sequential stream diff maximum = 0,
      Scheduling timeout scheme = 0,
      Timeout time/weight = 30,
      Timeout scheduling = 3,
      Scheduling priority scheme = 0,
      Priority scheduling = 3
   }, # end of Scheduler
   Use queueing in subsystem = 1
} # end of DRV0 spec

disksim_ctlr CTLR0 {
type = 3,
Scale for delays = 0.0,
Bulk sector transfer time = 0.0,
Maximum queue length = 2147483646,
Print stats =  1,
Scheduler = disksim_ioqueue {
     Scheduling policy = 1,
     Cylinder mapping strategy = 0,
     Write initiation delay = 0.0,
     Read initiation delay = 0.0,
     Sequential stream scheme = 0,
     Maximum concat size = 0,
     Overlapping request scheme = 0,
     Sequential stream diff maximum = 0,
     Scheduling timeout scheme = 0,
     Timeout time/weight = 6,
     Timeout scheduling = 4,
     Scheduling priority scheme = 0,
     Priority scheduling = 4
},  # end of ioqueue spec
Cache = disksim_cachemem {
     Cache size = 0,
     SLRU segments = [ 1.0 ],
     Line size = 64,
     Bit granularity = 1,
     Lock granularity = 64,
     Shared read locks = 1,
     Max request size = 512,
     Replacement policy = 2,
     Allocation policy = 0,
     Write scheme = 3,
     Flush policy = 0,
     Flush period = 0.0,
     Flush idle delay = -1.0,
     Flush max line cluster = 8,
     Read prefetch type  = 0,
     Write prefetch type = 0,
     Line-by-line fetches = 0,
     Max gather = 7
},  # end of cachemem spec
Max per-disk pending count = 16
} # end of CTLR0 spec

# system topology
topology disksim_iodriver driver0 [
     disksim_bus bus0 [
          disksim_ctlr ctlr0 [
               disksim_bus bus1 [
                    disksim_disk disk0 [],
                    disksim_disk disk1 []
               ] # end of bus1
          ] # end of ctlr0
     ] # end of bus0
] # end of system topology

disksim_logorg org0 {
   Addressing mode = Array,
   Distribution scheme = Striped,
   Redundancy scheme = Noredun,
   devices = [ disk0 .. disk1 ],
   Stripe unit  =  1,
   Synch writes for safety =  0,
   Number of copies =  2,
   Copy choice on read =  6,
   RMW vs. reconstruct =  0.5,
   Parity stripe unit =  64,
   Parity rotation type =  1,
   Time stamp interval =  0.000000,
   Time stamp start time =  60000.000000,
   Time stamp stop time =  10000000000.000000,
   Time stamp file name =  stamps
} # end of logorg org0 spec
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 20120226.parv
Type: application/octet-stream
Size: 9050 bytes
Desc: not available
URL: <http://lists.andrew.cmu.edu/pipermail/disksim-users/attachments/20120226/95e3cd50/attachment.obj>


More information about the Disksim-users mailing list