| Author |
Message |
babi
Guest
|
 Posted:
Thu Jun 30, 2005 4:17 pm Post subject:
Neural Networks implementation on SAN |
|
|
Dear friends,
Please let me know is there any recent work on SAN(Storage Area
Network) based on Neural Networks implementation.
Basically I want to know FC,SCSI,RAID or SAN which involved the
implementation on Neural Networks.
Thanks & Regds,
Babi |
|
| Back to top |
|
 |
Nik Simpson
Guest
|
| Posted:
Fri Jul 01, 2005 12:17 am Post subject:
Re: Neural Networks implementation on SAN |
|
|
"babi" <tmohanta@gmail.com> wrote in message
news:1120131842.847151.207260@o13g2000cwo.googlegroups.com...
| Quote: |
Dear friends,
Please let me know is there any recent work on SAN(Storage Area
Network) based on Neural Networks implementation.
Basically I want to know FC,SCSI,RAID or SAN which involved the
implementation on Neural Networks.
|
Huh? SANs deal with getting blocks from storage to hosts, as such it's hard
to see how (or for that matter why) you would need to implement a neural
network in your SAN.
--
Nik Simpson |
|
| Back to top |
|
|
babi
Guest
|
| Posted:
Fri Jul 01, 2005 3:35 pm Post subject:
Re: Neural Networks implementation on SAN |
|
|
Neural network can be used to perfomance prediction since the media
devices are involved. From the host system the intelligence can be used
to predict the intense I/O storage to route the request from host to
storage. This may be intereesting but I am not sure.
Thanks & Regds,
Babi |
|
| Back to top |
|
|
Others
Guest
|
| Posted:
Sat Jul 02, 2005 10:30 pm Post subject:
Re: Neural Networks implementation on SAN |
|
|
"babi" <tmohanta@gmail.com> wrote in message
news:1120214135.972078.29820@g43g2000cwa.googlegroups.com...
| Quote: | Neural network can be used to perfomance prediction since the media
devices are involved. From the host system the intelligence can be used
to predict the intense I/O storage to route the request from host to
storage. This may be intereesting but I am not sure.
Thanks & Regds,
Babi
|
My Only thoughts on this matter are that if you are using a neural net to
perform calcutions made up of various cells like grid network then a SAN
would assist in providing a storage network enabling data sharing through
SAN equivelant software to NFS, like SANFS, SANergy, etc..
Regards
Paresh |
|
| Back to top |
|
|
John Rigler
Guest
|
| Posted:
Mon Jul 04, 2005 6:27 am Post subject:
Re: Neural Networks implementation on SAN |
|
|
What problems are you trying to solve here? You may be on the road to
creating something too complicated to be reliable. I know that one thing
that could potentially be good about a SAN might be striping a little bit of
high-i/o data across a bunch of disks that are mostly full of data that is
rarely if ever read, this way you could get most of your heads spinning at
the same time, while not wasting gigs of space to do it. I suspect that you
are going after a certain level of flexibility, but if your solution is too
big you could be headed for the weeds.
c ya
"babi" <tmohanta@gmail.com> wrote in message
news:1120214135.972078.29820@g43g2000cwa.googlegroups.com...
| Quote: | Neural network can be used to perfomance prediction since the media
devices are involved. From the host system the intelligence can be used
to predict the intense I/O storage to route the request from host to
storage. This may be intereesting but I am not sure.
Thanks & Regds,
Babi
|
|
|
| Back to top |
|
|
Guest
|
| Posted:
Tue Jul 05, 2005 12:16 am Post subject:
Re: Neural Networks implementation on SAN |
|
|
In article <1120214135.972078.29820@g43g2000cwa.googlegroups.com>,
babi <tmohanta@gmail.com> wrote:
| Quote: | Neural network can be used to perfomance prediction since the media
devices are involved. From the host system the intelligence can be used
to predict the intense I/O storage to route the request from host to
storage. This may be intereesting but I am not sure.
|
If I understand this post correctly, you propose the following:
1. A. Predict performance of storage devices and SANs. This requires
a detailed understanding of the devices, the SAN components, and
the workload (the set of IOs that the application will issue).
B. Do this performance prediction using neural networks.
2. Then use the performance prediction to configure the workload and
SAN, optimizing it for your desired goals (low cost, good
performance, high reliability, certain tradeoffs between these
quantities, or whatever tickles your fancy).
These ideas are very old, and this path is extremely well-trodden. If
I were funny and cynical, I would say that this path is lined with the
dead bodies of hundreds of researchers and industry practitioners who
attempted to do this (and mostly failed, or died of old age before
succeeding). If you succeeded in this endeavor, you would have solved
one large problem on the way to building self-managing storage.
I've never heard of anyone using neural networks for the performance
prediction (the step 1.B. from above). The usual techniques use
ingredients from queueing theory, markov models, utilization metrics,
grey- and block-box techniques, heuristics. Also tried were genetic
algorithms and a few other exotics (most of which I've forgotten
about). There is maybe a half dozen textbooks on performance
prediction and optimization as applied to computer systems; most of
the stuff written in them has been applied to storage systems and
storage networks at some time or another.
Read the research literature, to get a level set on the state of the
art today. Easiest thing to do is to do a google search for the
"Minerva" and "Hippodrome" papers (authors include Wilkes, Merchant,
Alvarez, Veitch, and a host of others), and use those as a starting
point, following their references and citations (those two papers see
the world through one particular set of rose-colored glasses, so don't
assume that their conclusions are not controversial, because other
workers in this field have differently rose-colored glasses). Once
you are done with that, I would get my hands on the proceedings of the
three or four "FAST" conferences (stands for "File and Storage
Technologies", look at the Usenix website or in a library near you),
where you can study the state of the art in storage performance
prediction and self-management. For your particular interest, a long
look at the proceedings of the "SigMetrics" conference might also be
highly interesting.
--
The address in the header is invalid for obvious reasons. Please
reconstruct the address from the information below (look for _).
Ralph Becker-Szendy _firstname_@lr_dot_los-gatos_dot_ca.us |
|
| Back to top |
|
|
|
|
|
|
FAQ
Memberlist
Register
Profile
Log in to check your private messages
Log in