SAP Tech Ed 2006 – Day One impressions

2006-10-19I started off Tech Ed by going to Shai’s keynote, then squeezed in four sessions, a chat with an exhibitor and a fruitless search for an exhibitor who wasn’t there.
Keynote – Shai Agassi

  • Things that must be in place – foundations
  • Modernise the core
  • Optimise Business Usage
  • Drive strategic innovation Session: Improving “Portal over a WAN” Performance
    Exhibitor: VMware
    Session: Advanced Portal Infrastructure
    Session: Netweaver Mobile Roadmap
    No Exhibitor: RSA
    Session: Duet

    Beware: The following are hastily-typed stream-of-consciousness notes written in 15-minute spaces between sessions.

    Keynote – Shai Agassi

    The first event in Tech Ed 2006 was Shai Agassi’s keynote speech. This was scheduled to start at 8.45, which was a pretty early start for me – being an hour ahead of my usual timezone in the UK. I got to the RAI conference centre at 8.30, and after a pleasantly quick registration I made my way through an ever larger group of halls to the enormous Hall 8, beyond which was a humongous arena with a big stage at one end with two enormous Powerpoint slides projected either side of a pretty large video projection of the podium. I have to stop describing it there, because I’m running out of synonyms for “big”. At about 9am, Shai came on stage – just as well this was 15 minutes later than advertised, as it took that long just to get from the RAI entrance hall to the arena.

    The whole keynote was based on the metaphor of building a 2-floor house with a penthouse suite. There were four stages to business nirvana through Enterprise Services – setting the foundations, building the ground floor, adding a second floor, and finally enhancing the value of the place with a nice penthouse on top.

    Things that must be in place – foundations

    Basically, getting the foundations in place boils down to two things:

    • There must be one version of the truth (i.e. everyone should implement MDM)
    • Servers and processes should be consolidated (i.e. everyone should have an adaptive computing infrastructure)

    What’s interesting is that Shai seemed to be saying that if you’re not here, you’re nowhere. Don’t even think about doing any of the stuff that follows without doing this.
    I’m not sure I agree with the detail of what he’s saying, although I agree with the principles. One version of the truth is good, and that is why most people implement SAP in the first place – but I’m not sure that means having MDM. In fact, as far as I know, none of our customers uses MDM – but they would all say there is one version of the truth. This is just a matter of picking a “master” system for each set of data – it doesn’t mean that all of that data has to be in the same system. For example, I think that Active Directory is a perfectly good master system for user’s email addresses, logon IDs and favourite printers. SAP HR is a good place for their names, phone numbers and bank details. These can be linked on user ID or email address quite easily, without the need for an MDM system.
    Similarly, consolidating servers makes good business sense – but even if every package is running on its own server, that doesn’t prevent the move to an Enterprise Services Architecture (ESA) – some might argue that knowing where the server (instead of it being virtualised all over the place) is makes ESA easier rather than harder.

    Modernise the core

    OK, so having set the foundations we now need to build the ground floor. Putting this in place gives you an ESA platform on which you can build business value.
    The ground floor means just one thing: Everyone needs to be on mySAP ERP2005. “Honest, it’ll be stable for 5 years. We’ll just release enhancement packs. (Yes, I know we said the same about Netweaver ’04, but…)”. Hmm. This will be a hard sell to the customers where we’ve been implementing ERP 2004 for the past year or two.
    There is some method in this, er, thinking. Shai quoted a CEO as saying “Once SAP is in, you’re allowed to touch the core ERP system once every 5 years – preferably on a Saturday.” Then the killer: “But my boss want me to innovate every quarter.”
    The idea is that ERP 2005 is service-enabled, and SAP are publishing more and more enterprise services from this core (this will be what many of the enhancement packs will involve). Making services available from core means core can remain stable while innovation happens on the SOA platform that results. So, you start with ERP 2005 on a NetWeaver ’04s platform. On top of this you add another layer of NetWeaver (04s? 7.10?) which is where you build all your ESA composite applications.

    Optimise Business Usage

    So, with the ground floor in place (your newly modernised ERP 2005 core), it’s time to add the first floor of service-enabled applications which the modern core makes possible. Three examples were given of how this new core has enabled a great acceleration in SAP’s delivery of new products and technology.

    1. New user interfaces

    There are now lots of ways to reject someone’s leave:

    • Portal – using the Universal Worklist, as usual
    • Project Muse – still Universal Worklist, but via SAP’s Flex-based “Business Browser”.
    • Yahoo desktop widgets – I guess Google desktop widgets could also work
    • Phone calls initiated from Workflow – seriously, that has potential. When a high priority work item arrives in your inbox, the system actually calls you and asks what you want to do about it – voice recognition and everything. Was this just smoke and mirrors?
    • Or you can just carry on using SAP GUI – in Windows, Java or Web flavours

    Amazingly, these options were all demonstrated “live” on stage, with Shai’s subordinate Jeff applying for leave and Shai rejecting all attempts.

    2. BI Accelerator

    This was quite an impressive demo. Up until now, BW queries have been against a database of aggregated warehoused data on disk. Being on disk, this can be slow for large datasets. Being large datasets, aggregation of data is used to get reasonable performance but at the expense of limiting the range of queries which can be performed.

    Enter the BI Accelerator from Hewlett Packard – this is a blade-scalable hardware plugin for BI. It is based on TREX search technology and holds indexes for the entire data warehouse in memory – so no database software, no disk usage, no aggregates required for queries. The result is an enormous speed-up on queries, and no limit on the type of query. Shai demonstrated a query on 1 billion records in 3 seconds, using a mere 48 processor cores on a stack of blades.

    OK, so the hardware cost here was a significant fraction of a million dollars, but in certain contexts that delivers good value for money compared to waiting overnight for data aggregation jobs to complete.

    3. Enterprise Search

    Enterprise search is SAP’s TREX on steroids – it can be used to search ERP data as well as documents and websites via the portal. A trial version of this is already available for download from http://sdn.sap.com/downloads. Not much was really said about this, other than that it is a good application of ESA – because everything in the ERP core is available through web services, an existing search engine such as TREX can simply access the ERP system over HTTP and index the resulting XML data.

    Drive strategic innovation

    So now we have our nice 2-story SAP building, it’s time to top it off with a penthouse. This turned out to boil down to just a couple of job descriptions.
    What is currently the CIO (Chief Information Officer) job now becomes 2 jobs:
    Chief Process Innovation Officer – owns the global process map. Global process owners (GPOs) report in – there usually seem to be about 7 GPOs in a corporation. This is where the meat is – creating new processes using an estblished set of services.
    Chief IT Officer – responsible for server consolidation and producing a single version of the truth. The savings made here help enable process innovation.

    Session: Improving “Portal over a WAN” Performance

    Since it’s becoming more common for our customers to host their SAP systems in remote data centres, this session was quite interesting for me.

    To start with, the presenters discussed the main factors which affect the performance of browser-based applications. Here are the bullet points:

    • Latency is the big killer – therefore you need to reduce server roundtrips, or their effect.
    • Multiple server round trips can come from opening a TCP socket and opening a SSL session – hence the important of Keep Alives in HTTP 1.1 keeping a socket open over multiple requests
    • HTTP 1.1 specifies only 2 parallel connections (HTTP 1.0 defaults to 4 in most cases), but has many other benefits, including keep-alive & compression.
    • “Request pipelining” exists on some browsers¹ (the presenters didn’t say which), and this can get around most of the latency and parallelism issues.

    Next up, there was some discussion of a great new gadget from SAP to help speed things up, called “Netweaver Apps Delivery over WAN” or NW ADoW for short. This is a software appliance pair (one on the client network  – the CFE & one on the server network – the SFE²) can give 10x speedup, mostly because of  a new dictionary based compression method. This works by looking for common words & phrases in the traffic flowing between the two networks, and coding these as short byte sequences – only then is the resulting traffic compressed by gzip. This is combined with the usual Squid-style cacheing proxy which helps reduce latency – once one person on the client network has requested an image or JavaScript file, it is in the client-side cache for everyone else to use. ADoW is currently in pilot, available on a customer-by-customer basis.

    Finally, there were some tips on making the most of your existing setup with some tuning:

    • Make sure to turn on the Optimised version of EPCF in productive systems.
    • HTTP analysis tools are available: HTTPWatch and HTTPLook are browser based. There is also WireShark (formerly known as Etherreal) for on-the-wire monitoring, but this makes it hard to check SSL traffic. My own personal favourite here is Charles (http://www.xk72.com/charles)
    • Solution Manager Diagnostics provides an httpproxy tool which can record all traffic in a browser for later uploading to SMD where you can get very good analysis of bandwidth, latency, etc. This looked very impressive indeed – both client-side and server-side data are analysed to give an analysis of how much latency there is, and you can even view the resulting traffic flow in a GANNT chart.

    ¹ Pipelining can be enabled in Firefox – see http://www.hackaday.com/2004/12/26/speed-up-firefox/
    ² CFE = Client Front End, SFE = Server Front End

    VMware

    In between sessions I popped over to the VMWare booth for a chat. I’d wanted to run a copy of TREX inside a VMWare instance to get around some support issues with a platform we were running at the time, but when I raised a customer message with SAP they said they didn’t support running produciton systems on VMWare, mostly because Microsoft would not offer operating system support under VMWare.
    Anyway, VMWare are now apparently negotiating with SAP over supportability. There are no technical issues – everything runs fine – this is purely a matter of there being official support.

    Session: Advanced Portal Infrastructure

    A lot of this session was to do with how to design a network for security – inner and outer DMZs, Apache reverse proxy as an application gateway, web dispatcher as a load balancer, etc.

    A major point was made about the use of multiple portals in a deployment, federated at the point of consumption. For example, there would be a seperate portal for BI, because of strict dependencies of BI & Portal on particular service packs. There would be another portal for XSS for similar reasons. There would then be a corporate KM portal which also consumes BI & ESS in federated mode. This can also help with stability & scalability – if the BI server needs restarting, it doesn’t bring down ESS.

    Usefully, it has been found in general that typically only 10-15% of named users are concurrent – this information can be used for sizing. The Quick Sizer is now much better for portal installations.

    Session: Netweaver Mobile Roadmap

    SAP still don’t quite get it, but they’re moving in the right direction. SAP definition of Mobile is “occasionally connected device”, primarily a laptop, so the focus is on desktop apps with a synchronisation engine. In future releases they will support J2ME CDC profile, which is a big step in the direction of recognising that to the rest of the world, “Mobile” means handheld device – primarily Blackberry, but also platforms such as Windows Mobile, Symbian, and of course J2ME running on these platforms.

    No RSA

    There is a Netweaver Partner Pavilion with various companies present – but no sign if RSA Security, unfortunately.
    It’s a shame – I wanted to discuss RSA Access Manager integration now IISProxy (their current recommendation) is no longer supported by SAP as of December.

    Session: Duet

    Given an Exchange implementation and an SAP ERP server, all you need to add in to implement Duet is:

    1. Client add-in to Outlook
    2. A NW04 Java Add-on for the ERP system
    3. A new server, containing another NW04 J2EE stack to run SAP Duet services. This must be a Windows server, since it must also run MS Duet services.

    The HP guy next to me was very happy at all the extra servers needed.
    You can implement Duet in only 8 weeks (48 days – must be 6-day weeks) – so long as you’re surrounded by all the right experts (SAP, Microsoft, Duet) and infrastructure to start with.
    Currently there are four scenarios:

    1. Use Outlook calendar for time recording
    2. Have BI reports delivered to your inbox
    3. Create leave requests in Outlook calendar
    4. Integrate employee information from SAP HR into the Outlook Contacts
  • Leave a Reply