What could be the performance of my cluster
ctierney at hpti.com
Fri Apr 12 13:15:52 EDT 2002
All my experience is with oceanography and
Is the BLAST code something that spends lots
of time trying doing lots of little calculations,
or doing one big calculation? How important is
the speed of access to the database? What is
the memory footprint of the code when it runs
on the DS20E?
On Fri, Apr 12, 2002 at 03:32:34AM -0700, Suraj Peri wrote:
> Hi Craig,
> Many thanks for your mail. please excuse me for asking
> a dumb question and I am novice in this area.
> I am interested in using this cluster for BLAST
> purposes. I want to store ESTs( Expressed Sequence
> Tags) and GenBank ( nucleotide sequence database) and
> GenPept ( Protein sequence database) and total
> predicted protein sets of Human genome.
> I will use BLAST ( basic local alignment search tool
> algorithm) on this cluster. As the computataions are
> intensive and time consuming.
> So I wanted to compare the AlphaServer DS20E and my
> cluster in their computing abilities.
> Because there are no one in my friend circles no about
> this. Please help me if you have used clusters for
> BLAST purpose.
> --- Craig Tierney <ctierney at hpti.com> wrote:
> > It depends on what you are trying to do (doesn't
> > everyone
> > love that answer).
> > The number of flops your cluster can do should
> > be equal to:
> > flops = (no. of cpus) * (Mhz) * (flops per hz)
> > So for your cluster
> > flops = 8 * 1.53 Ghz * 2
> > I am assuming that with SSE you can get 2 flops
> > per cycle.
> > flops = 24.48 Gflops
> > Now, there are some issues with this. First, you
> > are never
> > going to get 1.53*2 Gflops out of a single
> > processor. Second,
> > leveraging all 8 cpus to get their maximum is going
> > to be
> > difficult if there is any communication between the
> > nodes.
> > Compilers play a big role in extracting the best
> > performance
> > out of the system. If you don't have a commerical
> > compiler
> > from the likes of Intel or Portland Group, I highly
> > recommend
> > getting one. You only have to purchase the compiler
> > for where
> > you compile, and not where you run. You can get
> > away with
> > one copy of the compiler on your server.
> > If you are trying to compare the AMD system to the
> > DS20E system,
> > it will depend on what you are actually trying to
> > do. If
> > you are running single precision floating point
> > codes that do
> > not require all the memory bandwidth a DS20E
> > provides, I would
> > think that within 10% that AMD processor will do the
> > work
> > of one 833 Mhz Alpha Cpu (You didn't say if you had
> > 2 cpus
> > in your DS20e). At least this is what I am seeing
> > for my codes when comparing Dual Xeon's, Dual AMD's,
> > and
> > dual API 833 boxes.
> > Craig
> > On Sat, Apr 06, 2002 at 03:35:45AM -0800, Suraj Peri
> > wrote:
> > > Hi group,
> > > I was calculating the performance of my cluster.
> > The
> > > features are
> > >
> > > 1. 8 nodes
> > > 2. Processor: AMD Athlon XP 1800+
> > > 3. 8 CPUs
> > > 4. 8*1.5 GB DDR RAM
> > > 5. 1 Server with 2 processorts with AMD MP 1800+
> > and
> > > 2GB DDR RAM
> > >
> > > I calculated this to be 48 Mflops . Is this
> > correct ?
> > > if not, what is the correct performance of my
> > cluster.
> > > I also comparatively calculated that my cluster
> > would
> > > be 3 times faster than AlphaServer DS20E ( 833 MHz
> > > alpha 64 bit processor, 4 GB max memory)
> > >
> > > Is my calculation correct or wrong? please help me
> > > ASAP. thanks in advance.
> > >
> > > cheers
> > > suraj.
> > >
> > > =====
> > > PIL/BMB/SDU/DK
> > >
> > > __________________________________________________
> > > Do You Yahoo!?
> > > Yahoo! Tax Center - online filing with TurboTax
> > > http://taxes.yahoo.com/
> > > _______________________________________________
> > > Beowulf mailing list, Beowulf at beowulf.org
> > > To change your subscription (digest mode or
> > unsubscribe) visit
> > http://www.beowulf.org/mailman/listinfo/beowulf
> > --
> > Craig Tierney (ctierney at hpti.com)
> Do You Yahoo!?
> Yahoo! Tax Center - online filing with TurboTax
> Beowulf mailing list, Beowulf at beowulf.org
> To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
Craig Tierney (ctierney at hpti.com)
Beowulf mailing list, Beowulf at beowulf.org
To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
More information about the Beowulf