If you’re following me on Twitter, you probably noticed I am attending SNIA’s Software Developer’s Conference 2010 (SDC 2010) and the CIFS/SMB/SMB2 PlugFest this week.
The agenda for the event is published at http://www.snia.org/events/storage-developer2010/agenda2010.
Here are some random notes from the event, some of them taken from a few earlier tweets (some sessions are just mentioned, others have a few notes).
I’ll keep updating today and tomorrow.

PlugFest

  • 25 companies attending the PlugFest
  • The place was busy starting on Sunday, 09/19
  • Windows Storage Server 2008 R2 “Birds of a Feather” session today at 7:30 PM, after the PlugFest Open House
  • File Server Capacity Tool (FSCT) “Birds of a Feather” session today at 8:30 PM

Session: "The Long and Winding Road to SMB/CIFS specs" by Christopher Hertel and Jose Rivera (both from ubiqx)

Session: "MS-FSA: Describing wire visible behavior of MS File Systems" by Neal Christiansen, Microsoft

  • [MS-FSA] documents algorithms that would otherwise have to be described in many other documents like CIFS, SMB, SMB2
  • Available for download from http://msdn.microsoft.com/en-us/library/ff469524(PROT.10).aspx
  • [MS-FSSC] (File System Control Codes) is an important support document for MS-FSA and other documents
  • Available for download from http://msdn.microsoft.com/en-us/library/cc231987(PROT.13).aspx
  • Going over a few of the algoriths described in [MS-FSA]
  • Name tunneling: preserves attributes of the file in certain situations (from 8.3 name transition)
  • Alternate Data Stream: Some interesting behaviors. For instance, by default, an empty primary DS is created
  • Byte range locks. Interesting concept of zero length ranges. FSBO also has some more info on those
  • USN journal or change journal. Describes the location points where the journal is updated
  • Directory change notification. You can see the many trigger points, what flags are set
  • Oplocks. How they are granted and how the different types interact. FSBO also has some nice tables on this
  • Deleting files. Interesting details on delete-on-close and delete-pending states
  • Valid Data Length (VDL). Allows setting file size without writing zeroes to newly allocated space
  • Object GUID. File can be opened by ObjectID. Requirements for setting, retrieving
  • Extended Attributes (EA). Name value/pairs to storage metadata associated with a file
  • Reparse Points: An open can be redirected. Used by symbolic links. Mutually exclusive with EAs
  • Symbolic Links and Mount Points. Implemented as reparse points

Session: “Challenges of 10 Gb iSCSI/FCOE Array Certification” by Mahmoud Jibbe, LSI

Session: “Comparison btw Samba 3 and Likewise Lwiod SMB FileServers” by Steven Danneman, Isilon

Session: “Status of Clustered CIFS Using Samba” by Volker Lendecke, Samba Team / SerNet

Session: “Implementing SMB2 within Samba” by Jeremy Allison, Google

Session: “SMB2 Meeting Linux” by Steve French, IBM

Session: Windows Storage Server 2008 R2 BoF session

  • Covered the main differences between Windows Server 2008 R2 and Windows Storage Server 2008 R2
    • Microsoft iSCSI Software Target 3.3
    • Single Instance Storage (SIS)
    • Browser-based Remote Management using RDP
    • OOBE (Out-of-the-box Experience), including two-node cluster OOBE
  • Highlights of the main improvements in the Storage Server Components
  • For a recent webcast on WSS 2008 R2, check on http://www.msteched.com/2010/NorthAmerica/WSV323

File Server Capacity Tool (FSCT) BoF session today at 8:30 PM

Session: Welcome Remarks by Wayne Adams (SNIA Board Chairman) and Mark Carlson (SDC Committee Chairman)

Session: Clouds, Convergence and Consolidation by Henry Baltazar, The 451 Group. http://www.the451group.com/

  • Where is the innovation? Cloud Storage Service, Hybrid Cloud Storage (on premises cloud gateway appliance)
  • Hybrid Cloud Storage Examples: Bottomless NAS Gateway, Disaster Recovery Appliance, Primary Storage Replacement
  • Convergence of Storage/Server/Network into super silos.
  • FCoE right now only in limited use cases: Server to SAN connectivity "top-of-rack", Storage target conectivity.
  • Solid State Storage - the driver is to close the Storage Performance "gap". Spinning drives not improving fast enough.
  • SSD: Is MLC the key? Overcoming write endurance limitations. 3x more affordable than SLC. Enterprise ready?
  • Where to deploy SSD? Server (PCI-E cards), Storage (embedded in the array), Network (bump in the wire cache)
  • Storage vendors are now using more powerful CPUs and running apps on the storage systems themselves

Session: Long Term Information Retention by Sam Fineberg, HP

  • If we want to keep data for 50 or 100 years we are talking going beyond the life span of any storage technology
  • Disaster, human error, attacks, economic/organizational faults, media/hardware/software/format obsolescence, losing context
  • A good archive is always in motion. You can't stuck a tape in a vault and expect to read it in 50 years.
  • What's important in a book? The words in it, the paper it's made of, the political context of when it was written?
  • Storage formats. We can't predice the future. Key properties: Self contained, Self describing, Extensible.
  • SNIA's SIRF (Self Contained Information Retention Format) and Preservations Objects. http://www.snia.org/events/storage-developer2009/presentations/tuesday/Fineberg-Cohen_LTRFormat.pdf

Session: Solid State Storage Panel Discussion: “Is Solid State Storage Ready for Primetime?”
Tom Coughlin (Moderator), Coughlin Associates; Marty Czekalski, Seagate; Jim Handy, Objective Analysis; Dr. Easen Ho, Calypso Systems, Inc; Esther Spangler, Smart Modular

  • Tom Coughlin
    • Looking at cost trends for flash memory and hard disk drives.
    • Historic symbiotic relationship between flash and HDDs in consumer devices
    • Symbiotic combination of Flash memory and HDDs in computers
    • Calypso SSD Blind Survey 2010. Comparing SLC, MLC and HDD. http://calypsotesters.com/testing.html
    • Flash Use: consumer applications, USB sticks. LA new cache layer? Use some flash or more DRAM for caching?
    • Hetzler’s Chasm Cache Analysis (S. Hetzler, IBM). Two chasms: DRAM-SATA, SATA-Tape. http://www.caiss.org/docs/DinnerSeminar/TheStorageChasm20090205.pdf
    • Dual drive (tiering on your computer – Marvell, Hitachi, Toshiba) and hyrid HDD (Seagate) approaches
    • Dual drive: How much SDD do you need to storage your applications? 40GB is enough. User data goes to a HDD.
    • Hybrid drives, totally transparent to the OS. Putting items you access more often on the SSD. Boot gets faster multiple reboots.
    • Options: Flash Cache Module, Solid State Drive, Solid State Hybrid Drive.
    • Hytachi-LG HyDrive – Optical disk with flash memory drive for dual storage laptops. http://www.engadget.com/2010/05/31/hitachi-lg-goes-official-with-hydrive-ssd-equipped-optical-driv/
    • HDD: Latency issues. Uses DRAM or flash memory for cache? Intel likes flash and hard drives. Seagate likes hybrid drives.
  • Jim Handy
    • The DRAM/HDD Speed Gap chart: Tape, HDD, DRAM, L3, L2, L1 – Price per GB vs. Bandwidth in MBps (logarithmic scale)
    • Options: Keep more data in DRAM (more RAM, more servers) or use accelerated HDDs to fill the gap ($300, short-stroked)
    • In 2004 NAND shot past DRAM’s price per GB. 1/12th the cost now and growing. NAND is slower/cheaper than DRAM and faster than a HDD.
    • Yesterday, Violin announced a 40TB SSD. “Put the whole database into flash!” $16 per GB and $3 per IOPS. http://www.violin-memory.com/news/press-releases/violin-memory-quadruples-data-center-density-with-40tb-capacity-flash-memory-array/
    • Automatic Tier Management: autonomous (controllers) or operating system support. Many more to come!
    • Performance muse be understood. Performance is not always intuitive
    • PC: 1 HDD, no hierarchy, speed is nice to have. Data Center: thousands of HDD, storage hierarchy, speed is money
    • PC architectures will follow the data center example. Main storage cached by flash. It’s going to happen, it’s just a matter of when.
  • Panel discussion

Session: Interoperability Tools for CIFS/SMB/SMB2 by Paul Long and Simon Sun, Microsoft

  • [MS-FSSO] is great place for an overview of file-related protocols. Great place to start.
  • You can also use the MSDN support forums
  • Network Monitor
    • Parsers updated monthly. You can modify/add parsers.
    • Conversation tree is nice.
    • Rich filtering and colorization of frames.
    • Frame annotations
  • Network monitor. Four sources of captures
    • GUI Capture from NDIS capture driver
    • NMcap / command line tool to enable/disable capture from a batch file
    • Pcap file format – import captures from other tools like tcpdump, wireshark
    • Event Tracing for Windows. Using netsh trace, logman. Can capture during reboot.
  • Netmon API. Can be used for capture or parsing (what is the SMB command in this frame?)
  • Experts / Top users / Top protocols / SSL decrypt / ...
  • Interoperability Test Suite
  • Captures to help you understand with annotations, reference to doc pointer - http://sysdoccap.codeplex.com 
  • Look for the file sharing system docs
  • Why reviewing packets against the protocol specification if netmon can do the first pass for you?
  • Parsers error will point to frames with issues. Also look at SMBerrors
  • There cloud be errors in the parsers, but most of the time…
  • Can also help you locate the protocol document doc based on a trace. Option to Go to data type definition
  • Spec explorer: Use a model to create tests: start with spec, create implementation from spec, write tests to check conformance
  • Model used to generate tests – provide input and check expected output
  • Hand create model, not test: Spec explorer handles model/input/output
  • Interop test suite:
  • Netmon vs wireshark
    • Netmon: parsers are very complete, focus on applications, can capture multiple NICs
    • Wireshark: more feature rich in general,  more focused on lower layers
  • Interop test suite: Intent to release source code in a VHD in the future. No specific timeline yet. Working on it.

Session: SMB Traffic Analyzer - Monitor the Data Flow of Samba Servers, Holger Hetterich, Novell

Session: Evaluating SMB2 Performance for Home Directory Workloads by David Kruse and Dan Lovinger, Microsoft

  • Options: Performance counters, network captures, windows perform ance analysis tool
  • Answering deeper questions. Am I storage, network or CPU bound? Am I seeing a high level of lock contention?
  • Event-based model – Log Request, Gather execution info, Log Response + Execution Info
  • Execution slicing
  • Data mining
  • Analyzing an FSCT run to look at local handle count across time, seconds per operation for each type of request
  • Evaluating the HomeFolders workload of the File Server Capacity Tool (FSCT)
  • Describing FSCT. Overview of the tool, history of its development, details about how it simulates a workload.
  • Homefolders: User follows a frequency distribution of scenarios, with ~1 scenario every 11 seconds.
  • Capacity metric: How many users the server can satisfy without overload (delayed scenarios)
  • Initial Physical Model: Client (SMB RDR), SMB Server, Storage Stack, Disk Subsystem. Track what happens at each level.
  • Model with compounding: 1 packet with multiple operations. Latency equation taken that into consideration
  • Measuring HomeFolders: SMB 2.1 total wire operations per FSCT scenario.
  • SMB Investigation: Modeling of Directory Cache Improvements vs. Directory Leases. Second one seems much better.
  • Latency Estimates;
    • Use the model to extrapolate to real hardware.
    • FSCT used to load low-end reference server, captured counters at near overload.
    • High Variance. Some large tails on the CREATE operation
  • Scaled scenarios. Extrapolate from LAN to WAN by adding 2-20ms client-server network latency
  • These are purely investigations, but shows how a performance model like this can be useful.
    • Consider performance counters and events in your design process
    • New protocols are opportunities to revisit your performance infrastructure
    • More data = More insight = Better designs, earlier
    • Extrapolation models can be an interesting input to the design process

Session: Scaling Likewise-CIFS Beyond 50k Concurrent Connections on a Single Node, by Gerald Carter, Likewise Software

  • Likewise as defined by Gerald: “Interoperability platform for non-Microsoft clients and servers in Microsoft OS dominated networks.” http://www.likewiseopen.org
  • Gerald quotes on performance:
  • “No one cares about performance until it’s not there.”
  • “Workloads, much like people, are rarely objective.”

If you are registered for the event, get the decks from http://www.snia.org/events/storage-developer2010/presentation

For everyone else, the decks are typically made available to the public a few months after the event. I’ll make sure to blog about it when it happens.