Summary of "Use of BLOB data types"

From: Carol Dionne <carold_at_corp.logica.com>
Date: Wed, 17 Nov 1993 18:12:18 GMT
Message-ID: <carold-171193100244_at_carold.logica.co.uk>


Here is are all the responses I've received regarding my post. I've merged followup responses into the original post to save space.

Return-Path: <li_at_sbctri.sbc.com>
From: li_at_sbctri.sbc.com (William Li)
Message-ID: <9311161718.AA18115_at_sbctri.sbc.com> Subject: Use of BLOB data types
To: carold_at_corp.logica.com
Date: Tue, 16 Nov 93 11:17:02 CST
X-Mailer: ELM [version 2.3 PL6]



We've used BLOB data type for image and voice using SYBASE. The BLOBs are retrieved in chunks (i.e. 4K, 8K). Depending on the application, the retrieved
chunk can be processed immediately, therefore, giving the appearance of immediate response. For example, display the data on the screen immediately
as each chunk is received. In this case the time to display the data is slower
than the time to retrieve each chunk.

For data insertion, the same method is used.

For small amount of data, some applications retrieve the data into local temporary files. Then, work with the file once the retrieval is complete.

We have experienced no delay problems.

In the near future, we will be examing the storage and retrieval of data greater than 2G, possibly using CDs.

The images are using a SQLServer on a 486 machine, clients are DOS/Windows. Approximately 20 concurrent users.

The voice application is using SUN Sparc10 with Sybase. This application is
in
beta, with an initial anticipated future volume of hundreds of concurrent users. This application is to eliminate telephone fraud of unauthorized use
of
calling cards by matching voice prints.

sbctri is Southwestern Bell Technology Resources, Inc., a small subsidiary of
Southwestern Bell Corp.

Bill Li



Return-Path: <grimes_at_access.digex.net> From: Seth Grimes <grimes_at_access.digex.net> Message-ID: <199311152154.AA00973_at_access.digex.net> Subject: Re: Use of BLOB data types
To: carold_at_corp.logica.com (Dionne, Carol) Date: Mon, 15 Nov 1993 16:54:42 -0500 (EST) In-Reply-To: <2CE7CF22_at_yankee.logica.co.uk> from "Dionne, Carol" at Nov 15, 93

    10:47:00 am
X-Mailer: ELM [version 2.4 PL22]
Mime-Version: 1.0

Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Content-Length: 566
------------------------------------------------------------------------------
I've been doing some evaluation work with Blobs in both Oracle and Sybase
--
and in InterBase, which has superior Blob handling but which management
wouldn't support because of its small market share -- but no performance
data yet.  I'm storing arrays of numbers and not images or sound and my
Blobs tend to be relatively small, but I do get better performance out of
them because Blob storage saves a lot of disk space compared to storing
the data in conventional fields.  Maybe when I get some usable numbers
I'll post a note to the nets.

                        Seth, grimes_at_oecd.fr

==========================
Return-Path: <mvalente_at_draco.lnec.pt> Date: Thu, 11 Nov 1993 17:23:10 +0100 From: mvalente_at_draco.lnec.pt Message-ID: <9311111623.AA13051_at_draco.lnec.pt> To: carold_at_corp.logica.com (Carol Dionne) Subject: Re: Use of BLOB data types Newsgroups: comp.databases.oracle,comp.database.sybase,comp.database.informix Organization: LNEC, Lab. Nac. de Eng. Civil, Lisboa, Portugal X-Newsreader: TIN [version 1.1 PL8] ------------------------------------------------------------------------------ In Oracle its LONG RAW that you have to use. I built a document storage tool [Intermedium] using just this capabilty. I not only store data but I also use LONG RAws to store programs which are then used to view the files. The tool is called Intermedium. It was built on a IBM RS6000 and uses Oracle6 and Motif. I'm sending it to you on the next messages. Its uuencoded. About performamce: I didnt notice any problems. On the other hand I didn't have a test database with that much volume. Check out Intermedium and tell me what you think. C U! By(e) Mario Valente
==========================
Return-Path: <mdb528_at_alaska.et.byu.edu> Date: Wed, 10 Nov 93 15:35:13 -0700 From: mdb528_at_alaska.et.byu.edu (Matthew D. Bennett) Message-ID: <9311102235.AA27420_at_alaska.et.byu.edu> To: carold_at_corp.logica.com (Carol Dionne) Subject: Re: Use of BLOB data types ------------------------------------------------------------------------------ I did a lot of work with long raw datatypes with my last client using Oracle7. There is a lot of fiddling around that you have to do and it is not the easiest thing in the world to do. If you are trying to do an evaluation in order to find out whose database you really want to use, I would suggest talking to the vendors. Good luck, Matthew D. Bennett ---------------------- In response to your question about if I would do it again, yes. I got paid a lot of money to do it the first time. I do think you are better off using a file manager for anything over 2 MB, however. Let me know if you need any more help. Matthew D. Bennett
==========================
Return-Path: <jean_at_gso.SAIC.COM> Date: Wed, 10 Nov 93 16:46:55 PST From: jean_at_gso.SAIC.COM (Jean Anderson) Message-ID: <9311110046.AA00829_at_ratatosk> To: carold_at_corp.logica.com Subject: Re: Use of BLOB data types CC: jean_at_gso.SAIC.COM ------------------------------------------------------------------------------ Oracle long raw's are fully logged (even in Oracle 7). Rollback segments and redo logs have to be large enough to accomodate the largest long raw. All blob access has to be from C and you have to manage everything yourself. In my opinion, Oracle simply checked the blob box off on a list of features. I am told that Oracle 8 will have better blob support. Rather than use long raws, we store pointers in the database to external files. Sybase lets you bypass logging at the database level; we separate databases between logged and unlogged blob data. That gets us further, but blob management is still really thin; there aren't a lot of functions to work with the data. We have started working with Montage, a new object-relational database that has much better support for large and binary data. You can bypass logging at the transaction level. You can insert and remove objects, and select the "handles" interactively with SQL. You can also hook in display functions to select the data from SQL and have it be visually displayed. The C interface has functions for managing and seeking into the object (i.e., you don't have to read data from the very beginning). I have been working with seismic, climate, and linguistic data. I understand that Montage has an Image data type extension specifically for multimedia data. If you are interested, they can be reached at sales_at_montage.com. My main problem with ODBM's is the general lack of standards (query language, interface, capabilities). An ODBMS standard is emerging (the Sep-Oct issue of Object Magazine lays out some of it), but it comes too late for us. And when all is said and done, we still want to take advantage of relational capabilities. Montage and UniSql are both object-relational; in my view, we get the best of both worlds. I'm working closely with Montage, my dba counterpart from the East Coast has been working with UniSql. We're going to get together next week for a couple of days to compare notes. It seems that Montage is more extensible. If you're interested, I'll keep you posted. Montage Software, Inc. 2000 Powell Street Suite 1405 Emeryville, CA 94608 Phone 510-652-8000 fax 510-652-8050 sales_at_montage.com UniSQL, INc. 9390 Research II, Suite 200 Austin, Texax 78759-6544 (512)343-7297 regards, -jean jean_at_gso.saic.com
==========================
Return-Path: <GOGI_at_aster.si> Date: Fri, 12 Nov 1993 12:29:10 +0200 (MET) From: GOGI_at_aster.si (Mladen Gogala) Message-ID: <931112122910.697_at_aster.si> Subject: Re: Use of BLOB data types To: carold_at_corp.logica.com X-Vmsmail-To: smtp%"carold_at_corp.logica.com" ------------------------------------------------------------------------------ I had significant delays with loading RGB images in ORACLE7 7.0.12.2 on SGI Indigo2 (Irix 5.1.1),which I where resolved as follows: a) Use OCI ,do not even dream of precompilers.I recommend the routine named "oflng" for fetching long values. b) I have call mallopt which defines options for malloc&calloc.It can speed programs significantly.Study the man page for malloc. 3) Dump ORACLE's MTS interface and use 'F:',the fast pipe adapter. It is significantly faster then the usual one 'P:' The performance gain was up to 40% with large images (~85MB).Fast pipe driver takes the credit for the largest part of the improvement. MTS stands for "Multi Threaded Server",the new way of communication between user process and ORACLE.It isn't a separate product,it should be bundled with RDBMS.MTS appeared with version 7,while pipe driver,fast-pipe, and SQL*Net ver. 1.2 drivers belong to version 6.The difference is the following: in version 6 each connection to ORACLE resulted in creation of another process,so called shadow process which performed privileged database access routines on behalf of the user who created it.This means that there was 1-1 mapping between connections and shadow processes. MTS makes shadow processes sharable:more then one connection can use the same shadow process at the same time.This interface is still immature and unreliable,particularly when transferring bulk of data between user program and ORACLE.Exactly this is the case when you load the big RGB image into ORACLE.(I dont know much about video sizes,but they are probably even bigger).My advice is :wait for the version 7.0.15.4.x to use ORACLE for such a purpose. Hope it helps! Kindest regards! Tel : ++386 61 1683-511 Mladen Gogala FAX : ++386 61 1683-165 ASTER Ljubljana Email: gogi_at_aster.si Snail-mail: ASTER Ljubljana Nade Ovcakove 1 The above statements represent The Absolute Truth, 61000 Ljubljana and therefore have nothing to do with the opinions Slovenia of my employer.
==========================
Received on Wed Nov 17 1993 - 19:12:18 CET

Original text of this message