Time to build GenBank on a SunFire server

Michael Black mbblack at cstone.net
Fri Feb 13 09:26:06 EST 2004


A build time of a day I would be very happy with!  Our RAID is connected
directly to the Sun.  My concern now is really that the Sun's internal
10,000rpm fibrechannel drive also seems extremely slow.  Our problem seems
to be generally massive IO bottlenecks, not specific to just the SCSI-RAID

Cheers, Michael
Michael B. Black
Molecular Biology Computing Support
Univ. of Virginia, ITC-ACHS
Charlottesville, VA, USA
mblack at virginia.edu

On 2/13/04 8:26 AM, in article c0ij9r$mpn$1 at news.urz.uni-heidelberg.de,
"Karl-Heinz Glatting" <glatting at dkdz.de> wrote:

> Hi Michael,
> we use a Sunfire 880 to build GCG indices for GenBank and it takes
> something like a day
> or so when doing a bit of parallelization. The disks are not local on
> this machine but on
> a Linux fileserver. So I guess your problem has something to do with
> network connection
> between the SUN and your fileserver.
> best regards,
> Karl-Heinz
> Karl-Heinz Glatting
> Bioinformatics Service Group
> Department of Molecular Biophysics
> Deutsches Krebsforschungszentrum (DKFZ)
> (German Cancer Research Center)
> Heidelberg, Germany
> E-Mail: glatting at dkfz-heidelberg.de
> Michael Black wrote:
>> Hello all,
>> I have a question about GCG performance on a Sunfire v480 server,
>> specifically, time to build genbank.  Last month, it took nearly 22.5 days
>> to build GenBnak 139.0, and that is just insane.
>> First, the machine specs:
>> SunFire v480 - 2x900Mhz UltraSparc III CPUs, 10GB RAM
>> SCSI controllor (SunSwift PCI) - 33Mhz, 32bit, 20MBps
>> RAID - CI Design, ultraSCSI2, 12x10,000rpm 74GB disks, RAID 5 (with one hot
>> swap) configuration.
>> As a test, we've tried building just plant+primate+rodent divisions of
>> GenBank 139.0, on the RAID and on one of the Sun's internal disks (which is
>> FibreChannel).  That build, which is about 10.7GB of data, takes about
>> 137min (on average) to build on the RAID, and about 36min (average) on the
>> internal disk.  This is just running "genbanktogcg" with dbindexing on, and
>> the flat files on the same device as the GCG build.
>> For comparison, doing the same test build on a dual 800Mhz PIII RedHat box
>> (RAID is similar size, but might be 7200rpm disks, I'm not sure) - local
>> disk took about 38min average, and the RAID about 40min average.
>> Prior to getting the Sun, we were running GCG on a 4x180Mhz CPU SGI Origin
>> 200 with 512MB RAM, and with the same CI Design RAID hung off it.  The last
>> GenBank build on that machine was release 137, which took about 8 to 9 days
>> to complete.
>> I fell something is drastically wrong with our SunFire!
>> If anyone else running GCG on a Sun, and locally maintaining GenBank, has
>> any insights on these speeds, I'd appreciate hearing from you.
>> TIA, Michael Black
>> Univ. of Virginia
>> mblack at virginia.edu

More information about the Info-gcg mailing list

Send comments to us at biosci-help [At] net.bio.net