Description of the changes for release 1.8b of LISTSERV(TM)
-----------------------------------------------------------
Copyright 1995 L-Soft international, Inc.
May 14th, 1995
*************************************************************************
**************************** Executive notes ****************************
*************************************************************************
The release notes for version 1.8b of LISTSERV(TM) have been split into
three documents: executive notes, list owner's notes, and LISTSERV
maintainer's notes. The present executive notes contains a brief
description of the most significant changes in version 1.8b, information
on new supported systems, new products and new licensing options, and a
Product Development report. For more information about any of the changes
described in the release notes, please write to [log in to unmask] or call
us at +1 (301) 942-8886.
**************
* Highlights *
**************
- LISTSERV currently ported to 11 major unix(R) brands, OpenVMS(TM),
Windows NT(TM) and Windows 95(TM)(*).
- LISTSERV-TCP/IP product for VM allows a smooth migration from BITNET to
TCP/IP on the same hardware. The migration procedure is very simple and
takes about 4 hours for 100 lists.
- Graduated and Limited licenses (now available for VM) significantly
reduce cost of ownership and TCP/IP migration for smaller sites.
- "SMTP controller" paradigm for VM sites: offload the resource intensive
SMTP deliveries from your VM system to an inexpensive "SMTP controller"
machine, transparently to your users. Hardware and operating system
costs for a workload of up to 250,000 daily deliveries (weekday
average) can be as low as US$6,000, with yearly license and maintenance
costs of US$550 only (academic/US prices; hardware prices may vary in
other countries).
- New LISTSERV High Performance product ("LISTSERV-HPO") for very large
workloads. LISTSERV-HPO can comfortably drive lists in excess of
100,000 subscribers with subsecond response times - on a personal
computer.
- New LSMTP(TM) product announced: expected to process 1-1.5 million
deliveries a day on a midrange workstation with 128M. Refer to separate
announcement for more information.
(*) For legal reasons, we may not sell the Windows 95(TM) version until
this operating system is officially released. The software, however,
has already been ported.
*************************************************************************
********************* Development Manager's message *********************
*************************************************************************
When we released version 1.8a of LISTSERV in December 1993, we were about
to embark on a very ambitious venture: free LISTSERV from its dependence
on mainframe hardware and on the NJE protocol, while preserving a high
degree of compatibility and interoperability with the existing LISTSERV
user base. Our original goal was to port LISTSERV to VMS(TM) and unix(R)
by 1Q95, and to Windows NT(TM) by 4Q95. This was no easy task to start
with, and with no industry precedent to support the feasibility of our
porting plan, which did not require a complete rewrite or involve POSIX,
there was serious concern among our customers that we would not succeed
in overcoming the difficulties that laid ahead of us. Well, we had our
doubts and crises, but we just grinded our teeth and pressed on. We had
to work double shift for several months, but we made it, and the results
far exceed our expectations.
Today, LISTSERV is available for both VMS(TM) architectures, for 11 major
unix(R) brands, for Windows NT(TM), for Windows 95(TM), and, of course,
for VM. A MPE/iX version should become available soon, and we are
considering a port to MVS and to the Macintosh(R). Many sites have
already migrated to the new versions of LISTSERV. The largest list in the
network, TOPTEN (over 73,000 subscribers), runs on a unix(R) LISTSERV.
The 4th largest LISTSERV site is also a unix(R) server, and probably the
largest unix(R) list manager in the network with over 2.1 million
deliveries a week. We already support more unix(R) brands than some of
the main competing products, and our unix(R) products perform better than
systems that were specifically designed for unix(R) and that are not
portable. LISTSERV now delivers about 10 million messages worldwide on an
average business day, a threefold increase from the release of 1.8a in
4Q93. In fact, we expect these numbers to soar as the PC versions of
LISTSERV are released and as Internet service providers start offering
mailing lists to their customers.
Naturally, we do not intend to rest on our laurels. The porting effort
will continue throughout the year, as the database and some of the more
advanced file server functions still have to be ported to non-VM systems.
At the same time we will continue to improve the functionality and
performance of LISTSERV to better meet the needs of new user groups. For
instance, Internet service providers need to be able to run very large
numbers of lists, whereas many of our corporate customers are planning to
start electronic newsletters with target audiences in excess of 100,000
subscribers. Even though LISTSERV has always excelled in the area of
performance, it was not able to meet the 50,000 daily subscriptions mark
that some of our customers were expecting after a large advertisement
campaign - at least not on inexpensive hardware. With the new High
Performance version of LISTSERV, which is being released together with
version 1.8b, you can run such lists even on a personal computer, and
with sub-second response time.
We are also working on two new products that will make it easier and
cheaper for people to use LISTSERV. The first is a state of the art WWW
interface, with support for many advanced functions such as management of
list subscriptions on an individual user basis (similar to a graphical
user interface, but with the advantage that it can be used from any
operating system that supports WWW), and point-and-click list
reconfiguration for list owners. We do not have a firm release date yet,
but we have made good progress and expect that the beta will begin
3-4Q95.
Our second new product is a high performance SMTP delivery program,
called LSMTP(TM), which was specifically designed to handle very large
volumes of mail and require as little maintenance and attention as
possible, while implementing the latest SMTP protocol extensions. This
product was developed through a partnership with Altmayer & Associates, a
consulting firm headquartered in Bonn, Germany. LSMTP is already used in
production by two of our customers, and is expected to be able to handle
up to 1-1.5 million deliveries a day on a mid-range workstation with 128M
of memory. LSMTP is particularly well suited to large
announcement/newsletter lists because it handles massive message
submissions very well. In under a minute, it will have opened 1,000 SMTP
connections and will be delivering that many messages in parallel. Even
normal lists will benefit from this degree of parallelism, of course.
LSMTP is also ideally suited to Internet service providers, large campus
mail servers, etc.
Our goal and ambition is that, by 1Q96, it should be possible to run just
about any LISTSERV workload on hardware costing under $10,000, and with
point and click ease. Small businesses, K-12 schools, community colleges
and other "low budget" organizations should be able to purchase a normal
PC server, install LISTSERV and LSMTP, click through the configuration
screens, and announce their first mailing list 30 minutes later, without
ever having to worry about babysitting mail queues or getting patches for
the security exposures of the month. Our customers are demanding "install
and forget" solutions, and this will be one of our main areas of focus
for the next 12 months.
Eric Thomas
Manager of Design & Development
L-Soft international, Inc.
*************************************************************************
********************** BITNET to TCP/IP migration ***********************
*************************************************************************
Many of our customers, including some of the largest LISTSERV sites, have
successfully migrated to TCP/IP. As of 9505, 7 out of the top 20 sites in
terms of volume delivered were already running LISTSERV-TCP/IP, and 8 out
of the 20 largest LISTSERV lists were managed by TCP/IP servers. The
migration procedures are now ironed out, and with our new licensing
schemes for smaller sites and the research we have been doing on
inexpensive mail delivery servers, migration costs are lower than they
were a year ago. Perhaps the most visible outcome of this experience and
research is that, whereas we had about a dozen offers and solutions last
year, we can now handle 95% of cases with just two migration plans: one
for sites that have decided to phase out VM, and one for sites that have
long term plans for their mainframe systems.
Plan number 1: VM system to be phased out within 12-18 months
-------------------------------------------------------------
If you are planning to phase out your VM system, the only issue to be
resolved is which system you want to migrate to (unix(R), VMS(TM),
Windows NT(TM)), and when. In most cases these choices will already have
been dictated by strategic management decisions; if you are hesitating
between several possible operating systems, we will be pleased to assist
you.
In this scenario, you will need a LISTSERV license for the target
operating system, hardware on which to run it, and, possibly, a
maintenance contract for your existing VM license, depending on when
exactly you intend to migrate. For moderate workloads (50,000 daily
deliveries or less), you can usually run LISTSERV on an existing system.
For larger workloads we recommend a dedicated machine.
For your LISTSERV license, you have three options, depending on your
workload, growth plans, procurement policy, and so on:
1. You can purchase a perpetual, unlimited capacity license, which comes
with one year of warranty (new versions, bug fixes, technical
assistance). From year 2, you purchase continued maintenance on a
yearly basis, at your option. This is the traditional licensing
solution, which we recommend for large workloads.
2. Alternatively, you can rent a "graduated license", which includes a
limited capacity license and a maintenance contract (new versions, bug
fixes, technical assistance) for a yearly charge. You can purchase
capacity upgrades at any time, as your needs change. This is the best
solution for smaller workloads - you pay according to the number of
lists you are using, and there is no initial investment.
3. If you are operating on a small budget, you can purchase a perpetual,
limited capacity license without maintenance (these licenses come with
only three months of warranty). Later, as new versions are released,
you will have the options to purchase upgrade kits, and naturally you
will also be able to purchase capacity upgrades, even if you elect not
to purchase the new versions.
In most cases, the optimal licensing policy is a simple function of the
number of lists that you are running. On the other hand, some
organizations find it much easier to obtain one-time money than regular
funds for maintenance. In such cases, it may be more appropriate to
select the first option even if the workload does not warrant it. We can
also bundle up to 3 years of maintenance with options 1 or 3. Contact
[log in to unmask] for more information.
Plan number 2: long-term future for VM
--------------------------------------
If VM is staying, the simplest solution is to keep all the lists, files
and other LISTSERV resources on VM, where your users expect them to be,
while offloading the resource intensive SMTP deliveries to an inexpensive
system (which you can view as a kind of "SMTP controller" for your
mainframe). Even on the very largest VM LISTSERV sites, LISTSERV only
consumes a few percent of one CPU. It is the SMTP servers that account
for the bulk of the mailing list cycles, and that do so much I/O that
each server essentially renders one volume unusable for other purposes.
There may be reasons to migrate LISTSERV to another machine of course,
but resource consumption is simply not one of them. By keeping LISTSERV
on VM, you simplify the migration considerably - for your technical
staff, for your users, and even for your operators:
- Because the lists and other LISTSERV resources remain on VM, the change
is totally transparent to the users. Only the most technical users will
notice the extra "Received:" line in mail headers. Others will just
think the mainframe was upgraded, tuned, or just had spinachs for
breakfast.
- Naturally, leaving the lists on VM also means significantly reduced
manpower costs for the migration. A simple, one-line configuration
change directs the deliveries to the "SMTP controller". You do not need
to rebuild the lists, recreate file catalogs and list archive
structures, etc.
- The mainframe, with its high reliability, 24h operations and
centralized, dependable backup procedures is the perfect place to keep
your mission critical data. Many PC and workstation systems, on the
other hand, use a file system where the integrity of your data is not
guaranteed; files may be lost or corrupted following a power outage or
system failure. In fact, this is the second largest source of incident
reports for the workstation versions of LISTSERV. There is no need for
you to go through any of that when you can simply keep the data on VM
where it is safe.
- The "SMTP controller", with its more limited RAS and backup procedures,
will not contain any permanent mission critical data. Because the only
customer data on the machine is the in-transit mail queue, you will
probably not need to back up the machine regularly (the SMTP mail queue
is similar to a RSCS/BITNET spool queue for a high speed link - the
files are processed so quickly that not even daily backups are of much
use). In a way, the SMTP controller is similar to an 8232/3172: you
back it up every time you install a new software revision, or then you
back up the media containing the software upgrade kit so that you can
reapply it on top of the original distributions if the worst should
happen.
- You can use the VM system as a backup while you repair/reinstall the
SMTP controller, whereas if LISTSERV were on the SMTP controller, you
would have no service when it goes down. SMTP controllers are so cheap
that you can even buy two for redundancy.
While L-Soft could sell you a turnkey SMTP controller, it is usually much
more cost effective for academic sites to build one themselves, taking
advantage of the generous academic discounts from most computer
manufacturers and available on-site expertise with non-mainframe systems.
Here is how you can implement your migration at the lowest possible cost.
You will need to purchase:
* The SMTP controller hardware and operating system (see below for some
possible hardware selections).
* A "DISTRIBUTE-only" LISTSERV license for the SMTP controller (typically
US$550/year including maintenance, but please contact [log in to unmask]
for a binding quote).
* A maintenance contract for your VM LISTSERV and/or LMail.
* A LISTSERV-TCP/IP license upgrade for your VM system, if you also want
to migrate from BITNET to TCP/IP. If LISTSERV is the only reason you
subscribe to BITNET, this is usually more cost effective than
continuing your BITNET membership. Switching to TCP/IP also insulates
you from the problems associated with the BITNET core saturation; in a
worst case scenario, the TCP/IP servers can be reconfigured to
completely ignore the existence of the overloaded BITNET servers, using
the Internet for routing and delivery. The BITNET servers, on the other
hand, are intrinsically captive as they can only operate through the
BITNET network.
Technically, the "SMTP controllers" can be any PC, workstation or
server-based system running any operating system with a SMTP
implementation that can handle your workload. In practice, the goal is to
save money, so you will want a PC or workstation based system. Since this
machine will have no interactive users and (as we have seen) no mission
critical data, the only aspects that really matter are price,
capacity/upgradability, and the general level of reliability of the
hardware and operating system.
Using PC technology, a SMTP controller with a sustained capacity of
360,000 daily deliveries (suitable for workloads with a weekday average
of 250,000 or less) can be had for as little as US$6,000 (US price from a
major, "serious" manufacturer with 3 years of warranty). In practice, you
will probably want to spend a bit more for a more robust design with
redundant fans and power supplies, ECC memory, additional SIMM slots for
upgradability, thermal monitoring, UPS, weekend and evening service
coverage, etc. A top of the line, high availability PC-based solution can
still be had for under US$10,000 - or you could just buy two of the
$6,000 systems for redundancy.
Of course, with academic discounts it may be more cost effective to use a
workstation system. We are only mentioning the PC based systems because
we have found that most VM customers assume that a server or at least a
high-end workstation system is required for workloads in excess of
150,000 per day. In fact a simple PC can handle much more than that. You
can migrate your workload off VM even if your total budget is only
US$10,000 (US estimate; hardware prices may vary in other countries).
*************************************************************************
***************** New LISTSERV High Performance product *****************
*************************************************************************
While LISTSERV has always had the reputation of being the fastest
available list manager, the democratization of the Internet is creating
new needs and placing new demands on Internet solutions - at a rate which
far exceeds the (already impressive) increases in computing power. Two
years ago, a "very large" mailing list was 15,000 subscribers, "a lot" of
lists meant a couple hundred, and "a whole bunch" of subscribers would be
a total of 50,000. Having been designed for scalability, LISTSERV was
able to handle these workloads very well. Today, a very large list would
be 75,000-100,000 subscribers, a lot of lists would mean 500-1000, and a
whole bunch of subscribers would be 100,000-200,000. LISTSERV can still
handle these workloads without problem.
Tomorrow's customers, however, are looking at much larger workloads.
About 6 months ago, a large corporate customer asked us if LISTSERV could
handle 50,000 daily subscriptions to a list of 100,000 or more
subscribers. They were planning to start electronic newsletters for some
new products, and wanted to advertise this Internet presence in the
press. Their marketing department estimated that there might be up to
50,000 subscriptions a day, and they wanted to know what kind of hardware
they would need for this. Naturally they wanted a scalable solution, so
it had to be possible to double the capacity for at most 3 times the
original cost. Well, LISTSERV could not do it. We looked at the fastest
non-mainframe systems available at the time, and found that LISTSERV
would barely meet the 50,000 requirement. Being the top of the line,
these systems would not be upgradable. With peered lists and a second
system, we might have been able to make it, but this was not the
customer's idea of a simple, robust, easy to maintain solution. Unless
you were willing to buy a Cray, this problem could not be solved by
buying faster hardware.
Many other corporate customers expressed a similar interest. Some were
expecting 500,000 to 1,000,000 subscribers in one year. Others had more
modest needs, but the postings were time critical and had to reach all
250,000 subscribers in an hour, barring a network outage. These were all
problems for which no solution existed, but for which there was a clear
business need. So, we decided to work on these problems, and the High
Performance version of LISTSERV (which we dubbed "LISTSERV-HPO") is one
of the solutions we have developed to meet these needs. Combined with
LSMTP, LISTSERV-HPO should be able to meet the needs of tomorrow's
customers.
But LISTSERV-HPO is not useful only to customers with millions of
subscribers and tens of thousands of daily subscriptions. If you have a
large but more modest workload, LISTSERV-HPO can help you decrease your
hardware costs. To give you an idea of what LISTSERV-HPO can do on modest
hardware, here are some figures from benchmarks made on one of our LAN
servers - a 90MHz Pentium(R) with 32M of (non EDO) memory, running
Windows NT(TM). This is the kind of system that most people buy for their
LAN services nowadays.
After creating a list with 100,000 subscribers, we added 100 new
subscribers (with the QUIET option, to measure the performance of
LISTSERV and not that of the mail system). The elapsed time, for all
100 users, was 2.2 seconds. That is, the average ADD time per user was
0.022 second elapsed. Naturally, in a real world case you would not use
the QUIET option, and it would take a few seconds to a few minutes for
the users to receive the welcome message. But these are fixed resource
costs which do not depend on the size of the list. The time spent
actually adding the user to the list is 0.022 second - with 100,000
existing subscribers. The mail delivery tasks scale up very well to
multiprocessor systems, and are the kind of problem that can be solved
easily by purchasing more hardware.
Of course, you probably do not run lists with 100,000 subscribers, nor do
you need the ability to deliver millions of messages a day. You probably
have a few lists in the 5,000-20,000 range, generating around 250,000
daily deliveries (M-F) which historically have been doubling every year.
Your purchase must last at least a year, because this is your budgeting
cycle, so you need a machine capable of handling at least 500,000 daily
deliveries, and preferably 750,000 to avoid bad surprises. You were
planning on migrating this workload to a high-end workstation with two
processors, 192M of memory and 6 SCSI disks on two SCSI buses for good
swapping response time, with a list price on the order of $50,000. You
know that this configuration can do 250,000 daily deliveries easily,
because others have done it, and you can reasonably hope that it will
scale up to 500,000. Beyond that, there is no hard data, because no one
is running mailing list workloads generating more than around 450,000
deliveries on busy days (except on VM). In fact, there are a number of
known scalability problems with sendmail, and another known problem is
that, on a system with hundreds of sendmail processes and heavy swapping
rates, it may be difficult to make the LISTSERV process run smoothly and
not accumulate a backlog.
With LISTSERV-HPO and LSMTP, you could run your 500,000 daily workload on
a workstation costing around $20,000 list. For another $5,000, you could
add the necessary memory to allow LSMTP to use the system to its fullest
and reach an estimated 1-1.5 million daily deliveries (see the separate
announcement for more details on LSMTP's performance and planned
availability). You save $25-30,000, get more capacity, and the assurance
that you will not run into unexpected scalability problems as your load
increases. The reason this solution is so much cheaper is that LSMTP does
not create a new process for each connection, and thus requires a lot
less memory. This also means your machine does not swap, and you only
need to provide I/O bandwidth for the mail queue. You can handle 500,000
deliveries with just 64M (assuming the workstation is not used
interactively, or at least not with X-Windows), with the standard system
disk that comes with the machine and an additional high-performance disk
for the mail queue. To double your capacity, you simply double the amount
of memory from 64M to 128M.
*************************************************************************
******** Free LISTSERV-HPO upgrade for early VM TCP/IP customers ********
*************************************************************************
Customers who purchased our LTCP-9000 package last year (or its corporate
equivalent) will receive a free upgrade to the High Performance version
of LISTSERV-TCP/IP for their VM system. This upgrade includes a new
perpetual license for LISTSERV-HPO and a free maintenance upgrade for the
current maintenance term. At the end of the current maintenance term,
customers will have the option to renew their maintenance for the
LISTSERV-HPO product, or to revert to the regular LISTSERV product to
save on maintenance charges. For VM customers with large workloads, the
difference in maintenance charges is expected to be less than the cost of
the CPU cycles saved thanks to the enhanced algorithms of the High
Performance version.
-------------------------------------------------------------------------
LMAIL, L-SOFT, LISTSERV and LSMTP are trademarks of L-Soft international.
Unix is a registered trademark of UNIX Systems Laboratories, Inc.
OpenVMS and VMS are trademarks of Digital Equipment Corporation.
Windows and Windows NT are trademarks of Microsoft corporation.
Pentium is a trademark of Intel Corporation.
All other trademarks, both marked and not marked, are the property of
their respective owners.
-------------------------------------------------------------------------
|