| T.R | Title | User | Personal
Name | Date | Lines |
|---|
| 39.1 | looking forward to hearing about this product | BISTRO::WATSON | genius is 99% desperation | Wed Nov 04 1987 09:06 | 11 |
| > I am engaged in a competitive sales opportunity which involves TANDEM's
> new "distributed, SQL, relational" (?) database.
Does it have a name?
> If anyone can dispel these nasty rumors, and provide
> some truth for these claims, please do it here, or send mail...
Please reply here so that we can all see the news.
Andrew.
|
| 39.2 | DB2 and Nonstop-SQL - The REAL competition | DEBIT::BERENSON | Rdb/VMS - Number ONE on VAX | Fri Nov 06 1987 03:15 | 4 |
| It's called Nonstop-SQL. It is clearly the technology leader in
relational databases for OLTP.
They've demonstrated 200 Debit-Credit TPS.
|
| 39.3 | It's Real | 33834::BOOTH | A career of MISunderstanding | Fri Nov 06 1987 18:25 | 15 |
| More bad news. It is truly distributed across the multiple processors
in the Tandem environment. However, their TPS claims are based on
Debit-credit, which lends itself well to partitioning across
processors. Most real-world applications are not that flexible for
partitioning purposes. In other words the 200+ TPS will be very
difficult to attain with a real-life database.
Nonetheless, Mr. Berenson is quite right. Tandem has about a 12-18
month lead on the reast of the industry in relational capabilities.
What they don't have is any software that layers on the database.
As a standalone TP engine Nonstop-SQL is great. It is NOT, however,
one of a series of tightly integrated database products for the
Tandem operating system, which was rewritten to allow for optimization
of the database product.
---- Michael Booth
|
| 39.4 | Nonstop perplexity | SMEGIT::RYDER | Al Ryder, aquatic sanitary engineer | Wed Nov 11 1987 16:38 | 12 |
| re Note 39.3
"What they don't have is any software that layers on the database.
As a standalone TP engine Nonstop-SQL is great. It is NOT, however,
one of a series of tightly integrated database products for the
Tandem operating system ..."
Michael, I'm not sure I understand your comment. Are you saying
that Nonstop-SQL does not run under Guardian (or whatever they call
their Nonstop operating system), and so when Nonstop-SQL is running,
that installation at that hour is running nothing else?
Alton Ryder, who-is-just-now-learning-to-spell-database
|
| 39.5 | Clarification | AUNTB::BOOTH | A career of MISunderstanding | Fri Nov 13 1987 03:55 | 7 |
| What I meant was that Nonstop-SQL does not have a host of 4GLs,
report writers, or application generators that will layer on it.
Yes, it (Nonstop-SQL) layers on VLX (the operating system). But
there is not a wealth of software that layers on the database.
Nonetheless, Tandem's database product is years ahead of everyone
elses.
---- Michael Booth
|
| 39.6 | Some Tandem Info | THEBAY::WAKEMANLA | Tall Duck and Handsome | Thu Nov 19 1987 22:40 | 18 |
| About the "Complies with all of Codd's Rules", I saw C.J. Date on
the sixth, and he stated that no one has yet complied with all twelve
rules. He also stated that Ted Codd is in the process of redefining
the twelve rules and is currently up to 166.
The way to compete with NonStop SQL is the way you compete with
Tandem. The Tandem system has limited availability of application
software. The system is also very difficult to develop applications
for. I know of a Digital CMP that paid $250K to buy out of a Tandem
funded development effort, an S/CMP that is about to sue Tandem over
another project, and a local transit agency that has a Tandem project
that is now 5 years late and $26M over budget. Tandem also has
operational problems in that they require constant baby-sitting
at the system programmer level.
Hope this helps
Larry
|
| 39.7 | Tandem Timetable | AUNTB::BOOTH | A career of MISunderstanding | Fri Nov 20 1987 04:21 | 12 |
| I think we have some confusion here. Tandem's Nonstop SQL is
state-of-the-art. However, it is not yet released. .6 is right about
Tandem without Nonstop SQL. However, the consensus on the new operating
system and database is that they are excellent, and rectify many
of the problems Tandem had in the past. Nonetheless, Nonstop SQL
is not due to be released at all until Spring, 1988. Full production
will not begin until Summer 1988, Until then Tandem has problems.
There is also an article on Nonstop SQL development environment in
this week's MIS Week. That article and others indicate that with
the new release Tandem will turn the corner and be competitive
everywhere.
---- Michael Booth
|
| 39.8 | DIOMEDES, Distributed DBMS | CHEFS::MCDONALDA | | Fri Nov 27 1987 12:50 | 130 |
|
Will this see the light of day?
For the past 18 months (or so), I have been working as a
designer/consultant on the development of a Fully Distributed Replicated
Real-time Relational Database.
- The product is called DIOMEDES. It is being/has been developed
by a company called Software Sciences Ltd, partly as a private venture
and partly for the British Royal Navy. A number of prototypes
have been produced since 1978, to establish the validity of the designs.
DIAS (Distributed Information Architecture for Ships) was very much like
Rdb, in that it shares host processor resources. The later ICDS
(Integrated Communications and Database System) was moved to a back end
machine (which provided the necessary power to run the DBMS).
Information pertaining to DIAS and ICDS is in the public domain and I
have access to the various reports and papers.
- DIOMEDES is a back end distributed database machine. It has
standard interfaces to hosts (e.g. VAX, Intel 80386, etc) and Local
Area Networks (e.g. IEEE 802.5, FDDI II, etc) and peripherals
(e.g. SCSI). Designed for harsh and hostile environments, the spec
demands that the lose and recovery of a node should not impact on the
system. I do NOT have access to papers on DIOMEDES. Sales literature has
to be requested from Software Sciences Ltd.
- The approximate performance of an fully replicated eight node
(ie a copy of the database exists on all the nodes - though handling
of partial replication is built in for large configurations) system is:
PER NODE: 300 (peak) DB reads + 300 (peak) DB writes + 50 (peak)
outstanding transactions.
SYSTEM: 2000+ DB reads and writes.
LATENCY: 10 milliseconds.
INTEGRITY: To 5 levels. DB integrity within 10 milliseconds of
commit.
UPDATES: Guaranteed, or roll back, delivery with 2 Phase commit;
with a NON-LOCKING protocol.
MULTI NODE (SITE) MULTI TRANSACTION COMMIT
- DIOMEDES employs partitioning.
- To achieve these performance levels, the database machine utilises
a multi-processor architecture consisting of a loosely coupled
network of 12 (or 14) T800 Floating Point 32-bit Transputers,
each with approximately 3 Mbytes fast dynamic RAM. This yields
approximately 120 (140) MIPs and 18 (21) MFLOPS. The cost per
machine (dependant on degree of ruggedisation, disc capacity,
performance requirements, etc) is, approximately, $45K (good
commercial quality, 20 Mbyte disc, card set) to $200K (Full military
ruggedisation, independant cabinet, power supply, etc).
- In addition to the database being distributed across a LAN (the
dynamic unit of replication and distribution being a page), the pages
held local to a node are, currently, distributed across four of the
processors. Thus, it is possible to to perform data searches and
writes in parallel, either for the same or different requests.
In addition to the processors that actually manage the database
pages (ie directly writing/searching memory or disc based pages)
there are the remaining processors that handle management functions
of a DBMS (e.g. Reads, five way joins, transaction monitoring,
host/LAN communications, health monitoring, Event Notification,
etc). These functions also operate concurrnetly with each other.
- A 'unique' and useful feature is Event Notification. This facility
removes the need for applications to poll the database, searching
for a significant event. In response to a once only request (a
modified read request containing normal qualification criteria
and/or formulae), the DBMS will monitor updates and will notify
the application(s) if an event has satisfied the request. This is
a particularly powerful feature for applications such as Air
Traffic Control. Operators need only instruct the DBMS to signal
them if a) any aircraft gets within, say, 1 mile of another aircraft
and are converging, or, more straight forward, 2) if any aircraft
within a 5 mile radius get within 1 mile of another.
- In addition to the actual DBMS HW and SW design, there is to be
a development environment. This is to consist of Data Dictionaries,
Simulators based on ORACLE, down load and save testing, test
harnesses with interfaces to the host based simulators and target
system, Ada SQL preprocessor (interface to the database is via
Ada applications using SQL commands), etc. There was no mention of
developing any application generators.
- Although currently a memory based database, geared to handling
large quantities of real time (Radar) data and providing fast
response to encyclopedic data, the concept of DIOMEDES (of a fault
tolerant distributed database) is well suited to handling of a
very large ( 10s - 100s Giga Byte) databases.
- When I left Software Sciences, the status of DIOMEDES was this:
i) A version of DIOMEDES was to be developed by next year (Q?). Thorn
EMI (a large UK electronics company) is funding this development, but
only to the tune of about #300K ($500K). Quite alot more cash is
required to develop a full function DIOMEDES. I believe the
Functional and System design specs have been completed.
ii) The part being funded by the Royal Navy is currently in Project
Definition phase. This will be completed end of Q3 1988. If an order
is placed, then FULLY FUNCTIONAL PRODUCTION versions will be
delivered within, about, 18 months of placement of contract.
____________ ____________ ____________
| HOST | | HOST | | HOST |
============ ============ ============
| DIOMEDES | | DIOMEDES | | DIOMEDES |
------------ ------------ ------------
|| || ||
|--------------------------------------------------------------| LAN
|| || ||
------------ ------------ ------------
| DIOMEDES | | DIOMEDES | | DIOMEDES |
============ ============ ============
| HOST | | HOST | | HOST |
------------ ------------ ------------
HOST: VAX, IBM, Intel 80386 mulit-processors, etc; run applictions.
All DBMS functions off loaded to DIOMEDES.
DIOMEDES: Provides TRANSPARENT access to database. All users, or
applications appear to have single access to one database
(subject to authority). Gateways provided to other Networks.
Angus
|
| 39.9 | good and bad | COOKIE::JANORDBY | | Wed Mar 30 1988 22:57 | 11 |
| Although NonStop-SQL is a relative infant, it is not quite as helpless
as we are led to believe in previoius notes. There is a COBOL85
compiler, a screen painter, and a number of other tools. NonStop-SQL
and Guardian, the operating system are highly integrated. When Guardian
comes up so does NonStop SQL.
Although they have demonstrated a huge number of TPS on a highly
artificial benchmark, field reports seem to indicate that they have
a difficult time beating Rdb in a real life runoff.
Jamey Nordby
|
| 39.10 | Does Tandem Have It Now | NCVAX1::OLSON | T7 this is K8. Nuke em til they glow. | Tue Apr 12 1988 00:08 | 33 |
| What is the latest news on Tandems Non-Stop SQL. The last notes
in this conference were written in November, 1987.
Is the product shipping.
Does the product offer;
Data Independance- one logical database
No Reliance on a Central Site- all functions are distributed
Backup and recovery procedures- accounting for failures on the WAN
or LAN (or software, disk, etc, etc)
Continuous Operation- node going away has no affect on the applications
ability to process data.
Fragment Independance- store data close to the user
Tranparency- both user and programmer
Updates- the whole idea of two phased commits across a physically
dispersed logical database
Replication Independance- allows performance tuning by replicating
data close to the user with no affect on other operations.
Others may know this as;
Retrieval Transparency
Update Transparency
Schema Transparency
Performance Transparency
Transaction Transparency
Copy Transparency
Tool's Transparency
Buffaloed by Tandem in MPO
|
| 39.11 | Summer | QUILL::BOOTH | A Career in MISunderstanding | Wed Apr 13 1988 15:19 | 6 |
| The last I heard, NON-STOP SQL will be officially released sometime
in the May-June timeframe. As far as I can tell, it has everything
you have mentioned, except maybe replication independence. I'm not
sure about that property.
---- Michael Booth
|
| 39.12 | Can you tell I'm from Tandem? | CSTEAM::WADSWORTH | | Tue Jul 26 1988 17:35 | 20 |
| Tandem NonstopSQL was distributed to customers in a limited release
of the Guardian operating system Level B41. Tandem has had many
threading problems bringing together SQL and other products.
Originally, NonstopSQL was due to ship with C00. However, that
release was held up until mid august when C10 will be officially
released to customers (maybe). Nonstop SQL is fully geographically
independent, provides an active data dictionary, supports ansi
standard SQL, and ISO SQL, is fully integrated with existing Tandem
software, and can be interspersed with Encsribe on the same system.
Applciations can access both SQL and Enscribe files from the same
server program. Defines are supported so that the applications
can be written without concern for the final residence of the data
files. SQLCI supports quick queries and formated reports.
TMF ensures consistency of the database over geographic distribution.
If any more specific information is required contact
Kirby Wadsworth
Competitive Sales Support
296-4113
|
| 39.13 | high on TPS, High on disks | COOKIE::JANORDBY | | Thu Aug 18 1988 21:41 | 17 |
| One item to note about Tandem is that according tothe DataPro price
lists, disk storage can be as high as $45 - $50/MB. A certain amount
of disk storage (2.7GB) comes with each VLX node. However, in order
to do all the processing that a given node can do, additional storage
may be necessary. Such was the case with Top Gun benchmark specs.
With 3 32-processor nodes running 208 TPS, 45 GB of storage is required
to run Debit-Credit (220MB/TPS). Since only 8.1 GB was included
in the system as delivered, their configuration would cost an
additional $1.6M just to cover required disk capacity. They actually
scaled their benchmark to only use an additional 1.66 GB per node(total
of @5GB). So all this blabber means that Tandem disk prices are
extraordinarily high compared to IBM or DIGITAL. The more disk storage
that is required for a given system, the wores price performance
will be for Tandem.
Jamey Nordby
|
| 39.14 | RIP MIPs HEREs TPS | CSTEAM::WADSWORTH | KIRBY WADSWORTH | Mon Aug 22 1988 17:20 | 13 |
| Please keep in mind that Tandem relaxed the contraints of D/C to
reach the 208 tps number. Their true performance under the D/C
requirements is about 104tps. There is very general agreement from
industry analysts that this is a more true to test number.
It does not matter which constraints were relaxed, or what real
life applications the customer is building. D/C is not a test of
real life, it is an artificial test of machine power, just like
MIPS used to be. The only reason it is fairer than MIPS is because
it establishes a framework for testing like units of work. DEC could have
played around with the constraints until we found one or two that
would inflate our numbers, but we stuck to the rules and won, Tandem
did not.
|