Orbis2 Hardware Background
Presentation to MMG
by Audrey Novak
October 16, 2001
Orbis2 Hardware Background Review
Review our current Orbis Hardware environment and ITS's past reports on our Orbis2 Hardware Configuration Options.
Background Information from David Gewirtz about the preliminary hardware configuration proposal for Orbis2:
We run an IBM System 390 for Orbis at 155 Whitney. The S/390, tape subsystem, memory and for disk upgrades from 97-00 cost about $725,000 in hardware, $115,000 annual software fees, and $50,000 support, services and maintenance. Over a 3 year period we spent about 1.2 million on Orbis hardware, services and maintenance.
I want to talk about our vulnerability for a second, but first let me remind you that our Orbis data – all of it bibl, auth, invoice, patron, vendor, transactions, etc – is extremely well protected through backups which are stored both on-site and off-site.
On the S/390 we have mirrored disk drives, which means if one drive fails we will not go down because there is a duplicate drive running. Otherwise, if a component of the mainframe fails we will go down.
If there is a fire at 155 Whitney, or if the location loses electricity (from both of its power feeds) we will go down. We are vulnerable to these unlikely disasters.
Are we comfortable with this exposure?
I’m not sure that we are comfortable with it, but the cost of creating a backup or redundant environment is very high. We'd either have to maintain an annual contract with, and high speed network connections to, a disaster recovery location somewhere in the region (where customers are sharing the expense of an unused, backup machine) or we'd need to buy and support a second mainframe. Both are very expensive propositions. The Library and ITS determined that the cost of this insurance was out of line with the risk of a long term catastrophic failure such as a fire or other act that destroys the physical environment at 155 Whitney Avenue. Electricity, should if fail, is restored quickly. We protect ourselves from a computer part failure by buying fast response service contracts.
As you all know, Orbis on the S/390 has as David said, "a near perfect record for reliability, availability and serviceability (RAS)." We just don't go down. We don't go down due to hardware problems, environmental problems or development, test and upgrade requirements. We want to achieve this same level of RAS on Orbis2.
Projected Server Costs from David Gewirtz (year 2000)
Note that network infrastructure, peripheral and tape subsystems costs were NOT reflected in these estimates.
Also, a year ago the MMG identified the following goals:
As you know from David's February and May reports, our hardware vendor of choice, SUN was about to announce a new line of servers that looked very attractive. This line went public in the spring - it's called the Sunfire. ITS focused their Orbis2 hardware investigation on these new servers because the Sunfires have faster processors, a longer life-cycle (new technology), are more fault tolerant, and can be divided into partitions (called domains) just like the S/390. With a Sunfire it is possible that we could buy one physical server (box), divide it up into 4 domains and distribute our applications over 2 of those domains while using the 3rd as a test server and the 4th as a development server. That is, we get 4 machines for the price of 1.
At the end of August Endeavor and Sun arranged a certification test for the Sunfire out at the Sun benchmarking center in California. Over a 3 day period, David and Endeavor staff tested Voyager applications on this machine, tested some of the new SUN utilities that allow data replication, and ran a few basic performance tests. As a result of these tests Endeavor will certify Voyager on the Sunfire (or any machine that uses the same Ultra Sparc 3 processor). Performance tests were extremely encouraging.
Complicating our Orbis2 hardware investigation in the spring, Sun announced yet another new server line - one based on the same processor that is employed in the Sunfire, but with fewer built-in redundancy features (e.g., no redundant back-plane, the main circuit board that e.t. plugs into, and no partitioning capabilities). These servers will go public next month, they are much less expensive than the Sunfire. Four of these new servers cost less than one Sunfire divided into 4 partitions.
Where are we now?Our goals have not changed, we want to:
Return to Orbis2 Implementation Site