Last week I asked if anyone could identify High Performance Disk Systems.
Thanks to all who responded.
The responses were as follows:
From:vandyk@roses.rockwell.com (R. Michael Van Dyk)
The power of Digital's Alpha AXP systems has now been matched by the
performance of the VELOCITOR I/O subsystem. This product combines very fast
storage controllers and disks to handle the demands of the Alpha AXP systems
and I/O intensive applications
You can find out about this using a WWW client like Mosaic to access
DEC's web server (http://www.digital.com/info/forms/search.html)
and search for VELOCITOR.
From: Chris Swanson <cds@sanjose.ssds.com>
I have used the "Pluto" (SA100) in the past (and will be again
next week). It seems as if transfer rates are sustainable from about
10 to 20 Mbytes/sec (Burst to close to 25 MBytes/sec). This is
full-duplex speed.
From: Brad - Walker <bwalker@shell.portal.com>
S
I just finished a high-speed SCSI board for Sun workstations. I was able
to sustain 18MB/sec. with no problem. The company that sells it is called
Antares Microsystems 408-370-7287.
From: john@texmemsys.com (John Marsh)
We make a high-performace memory system, called the SAM-200, which is
DRAM-based and so is very fast, and can hold up to 8 GBytes. Our
current interfaces to SUN computers is either VMEbus or SBus, and the
best we have seen is about 16 MBytes/second from the SBus interface.
We are about to start working on a faster SBus implementation, but
that, as they say, is stil vaporware.
We do have a HIPPI interface that runs at up to 98 MBytes/second, but
I do not know of a good HIPPI interface for the SUN that would not be
limited by going through an intermediate step, such as VMEbus. Our box
can go very fast; the limitation is usually the target bus, or computer
operating system overheads. If you have any possibility of volume,
then we might be willing to take on a custom or special interface for
your application. We are also multi-ported, so that might be of
interest if you need to share data.
From: ken@cpatl.com (Ken Burns)
To: mbecker@elf.igraph.com
Subject: disk IO
I'm fresh off a SPARCstorage class and can help with the numbers and
"facts" of the SSarray. The box is tentativily rated at 2000 IO's per second,
by comparision a standard SCSI disk is good for 50 to 100 IO's/second. The
box has six separate SCSI II fast/wide internal buses. The disks are currently
1 Gigabyte Seagate disks, so if you set up a striped 4 gigabyte slice you should
get 4 disks on a single partition. This is the "advertised" best setting for
the box. In just playing with the box on a SS1000 and two disks stripped I
was able to sustain 500 IO's on this file system will running tar to another
file system and dd to a third.
Something to note is that 25 Meg/second is full-duplex, it can be
written to, while also reading from the the "pipe." The next best box is
the DG Clarion which is SCSI II fast/wide into the box, but only SCSI II fast
internal can only achieve 20 Meg burst in a single direction. The difference
under load between these two boxes is probably a factor of 4 to 6 in favor of
the Sun Storage array.
From: gp310ad@prism.gatech.edu (ROBERT DUCKWORTH)
Best measured I've seen is 11meg/sec.
I haven't looked at the Sun unit though.
From: Birger.Wathne@vest.sdata.no (Birger A. Wathne)
But I don't think you can expect 25Mb/sec for sequential data access
from a single file... You would have to check directly with each
manufacturer for those kind of figures.
I guess 25Mb/sec is max throughput on the bus (upgradeable to 50Mb/sec),
but to reach this capacity, you may have to read from several files, etc...
I don't know enough about the cluster yet, but I'll have find out these things
some day.
As the cluster has 6 high-speed SCSI chains, you may still reach 25Mb/sec
continuous data rate on sequential access, but check with Sun for
this.
Fast-Wide-Differential SCSI should offer 20Mb/sec transfer rate.
Running software striping (DiskSuite) over several chains should theoretically
offer a lot of bandwidth. but are the disks able to read that fast? I would
guess you could stripe severl disks on the same chain without banging
your head against scsi bus throughput. At least with these fast buses.
You should even be able to run 2 or 3 ordinary fast scsi disks on each
ordinary fast scsi-2 chain. If you then add 3 scsi chains, and interleave
the data (using DiskSuite and striping), you may be able to get something
approaching 30 Mb/sec. But you will need to balance the number of disks
on each chain with the max sequential data transfer rate of each disk.
From: "HALL, BRYAN D." <HALLBD@wsmc-mis.af.mil>
I'd recommend against the SA1000 for big-block fast I/O. The reason
is that it does not support RAID 3, but only 0, 1, and 5. You will get
much quicker transfer rates from something like a Megadrive SCSI F&W
raid box set to level 3, with the high-speed baracuda 7200 rpm drives.
In otherwords, with the correct F&W controller (buffered) on the sun
side, you can actually get 20MB/sec continous.
This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:09:09 CDT