Beyond the File System Designing Large Scale File Storage and Serving Cal Henderson

Web 2.0 Expo, 17 April Hello!

Web 2.0 Expo, 17 April Big file systems? Too vague! What is a file system? What constitutes big? Some requirements would be nice

Web 2.0 Expo, 17 April Scalable Looking at storage and serving infrastructures 1

Web 2.0 Expo, 17 April Reliable Looking at redundancy, failure rates, on the fly changes 2

Web 2.0 Expo, 17 April Cheap Looking at upfront costs, TCO and lifetimes 3

Web 2.0 Expo, 17 April Four buckets Storage Serving BCP Cost

Web 2.0 Expo, 17 April Storage

Web 2.0 Expo, 17 April The storage stack File system Block protocol RAID Hardware ext, reiserFS, NTFS SCSI, SATA, FC Mirrors, Stripes Disks and stuff File protocol NFS, CIFS, SMB

Web 2.0 Expo, 17 April Hardware overview The storage scale InternalDASSANNAS LowerHigher

Web 2.0 Expo, 17 April Internal storage A disk in a computer –SCSI, IDE, SATA 4 disks in 1U is common 8 for half depth boxes

Web 2.0 Expo, 17 April DAS Direct attached storage Disk shelf, connected by SCSI/SATA HP MSA30 – 14 disks in 3U

Web 2.0 Expo, 17 April SAN Storage Area Network Dumb disk shelves Clients connect via a ‘fabric’ Fibre Channel, iSCSI, Infiniband –Low level protocols

Web 2.0 Expo, 17 April NAS Network Attached Storage Intelligent disk shelf Clients connect via a network NFS, SMB, CIFS –High level protocols

Web 2.0 Expo, 17 April Of course, it’s more confusing than that

Web 2.0 Expo, 17 April Meet the LUN Logical Unit Number A slice of storage space Originally for addressing a single drive: –c1t2d3 –Controller, Target, Disk (Slice) Now means a virtual partition/volume –LVM, Logical Volume Management

Web 2.0 Expo, 17 April NAS vs SAN With a SAN, a single host (initiator) owns a single LUN/volume With NAS, multiple hosts own a single LUN/volume NAS head – NAS access to a SAN

Web 2.0 Expo, 17 April SAN Advantages Virtualization within a SAN offers some nice features: Real-time LUN replication Transparent backup SAN booting for host replacement

Web 2.0 Expo, 17 April Some Practical Examples There are a lot of vendors Configurations vary Prices vary wildly Let’s look at a couple –Ones I happen to have experience with –Not an endorsement ;)

Web 2.0 Expo, 17 April NetApp Filers Heads and shelves, up to 500TB in 6 Cabs FC SAN with 1 or 2 NAS heads

Web 2.0 Expo, 17 April Isilon IQ 2U Nodes, 3-96 nodes/cluster, TB FC/InfiniBand SAN with NAS head on each node

Web 2.0 Expo, 17 April Scaling Vertical vs Horizontal

Web 2.0 Expo, 17 April Vertical scaling Get a bigger box Bigger disk(s) More disks Limited by current tech – size of each disk and total number in appliance

Web 2.0 Expo, 17 April Horizontal scaling Buy more boxes Add more servers/appliances Scales forever* *sort of

Web 2.0 Expo, 17 April Storage scaling approaches Four common models: Huge FS Physical nodes Virtual nodes Chunked space

Web 2.0 Expo, 17 April Huge FS Create one giant volume with growing space –Sun’s ZFS –Isilon IQ Expandable on-the-fly? Upper limits –Always limited somewhere

Web 2.0 Expo, 17 April Huge FS Pluses –Simple from the application side –Logically simple –Low administrative overhead Minuses –All your eggs in one basket –Hard to expand –Has an upper limit

Web 2.0 Expo, 17 April Physical nodes Application handles distribution to multiple physical nodes –Disks, Boxes, Appliances, whatever One ‘volume’ per node Each node acts by itself Expandable on-the-fly – add more nodes Scales forever

Web 2.0 Expo, 17 April Physical Nodes Pluses –Limitless expansion –Easy to expand –Unlikely to all fail at once Minuses –Many ‘mounts’ to manage –More administration

Web 2.0 Expo, 17 April Virtual nodes Application handles distribution to multiple virtual volumes, contained on multiple physical nodes Multiple volumes per node Flexible Expandable on-the-fly – add more nodes Scales forever

Web 2.0 Expo, 17 April Virtual Nodes Pluses –Limitless expansion –Easy to expand –Unlikely to all fail at once –Addressing is logical, not physical –Flexible volume sizing, consolidation Minuses –Many ‘mounts’ to manage –More administration

Web 2.0 Expo, 17 April Chunked space Storage layer writes parts of files to different physical nodes A higher-level RAID striping High performance for large files –read multiple parts simultaneously

Web 2.0 Expo, 17 April Chunked space Pluses –High performance –Limitless size Minuses –Conceptually complex –Can be hard to expand on the fly –Can’t manually poke it

Web 2.0 Expo, 17 April Real Life Case Studies

Web 2.0 Expo, 17 April GFS – Google File System Developed by … Google Proprietary Everything we know about it is based on talks they’ve given Designed to store huge files for fast access

Web 2.0 Expo, 17 April GFS – Google File System Single ‘Master’ node holds metadata –SPF – Shadow master allows warm swap Grid of ‘chunkservers’ –64bit filenames –64 MB file chunks

Web 2.0 Expo, 17 April GFS – Google File System 1(a)2(a) 1(b) Master

Web 2.0 Expo, 17 April GFS – Google File System Client reads metadata from master then file parts from multiple chunkservers Designed for big files (>100MB) Master server allocates access leases Replication is automatic and self repairing –Synchronously for atomicity

Web 2.0 Expo, 17 April GFS – Google File System Reading is fast (parallelizable) –But requires a lease Master server is required for all reads and writes

Web 2.0 Expo, 17 April MogileFS – OMG Files Developed by Danga / SixApart Open source Designed for scalable web app storage

Web 2.0 Expo, 17 April MogileFS – OMG Files Single metadata store (MySQL) –MySQL Cluster avoids SPF Multiple ‘tracker’ nodes locate files Multiple ‘storage’ nodes store files

Web 2.0 Expo, 17 April MogileFS – OMG Files Tracker MySQL

Web 2.0 Expo, 17 April MogileFS – OMG Files Replication of file ‘classes’ happens transparently Storage nodes are not mirrored – replication is piecemeal Reading and writing go through trackers, but are performed directly upon storage nodes

Web 2.0 Expo, 17 April Flickr File System Developed by Flickr Proprietary Designed for very large scalable web app storage

Web 2.0 Expo, 17 April Flickr File System No metadata store –Deal with it yourself Multiple ‘StorageMaster’ nodes Multiple storage nodes with virtual volumes

Web 2.0 Expo, 17 April Flickr File System SM

Web 2.0 Expo, 17 April Flickr File System Metadata stored by app –Just a virtual volume number –App chooses a path Virtual nodes are mirrored –Locally and remotely Reading is done directly from nodes

Web 2.0 Expo, 17 April Flickr File System StorageMaster nodes only used for write operations Reading and writing can scale separately

Web 2.0 Expo, 17 April Amazon S3 A big disk in the sky Multiple ‘buckets’ Files have user-defined keys Data + metadata

Web 2.0 Expo, 17 April Amazon S3 ServersAmazon

Web 2.0 Expo, 17 April Amazon S3 ServersAmazon Users

Web 2.0 Expo, 17 April The cost Fixed price, by the GB Store: $0.15 per GB per month Serve: $0.20 per GB

Web 2.0 Expo, 17 April The cost S3

Web 2.0 Expo, 17 April The cost S3 Regular Bandwidth

Web 2.0 Expo, 17 April End costs ~$2k to store 1TB for a year ~$63 a month for 1Mb ~$65k a month for 1Gb

Web 2.0 Expo, 17 April Serving

Web 2.0 Expo, 17 April Serving files Serving files is easy! ApacheDisk

Web 2.0 Expo, 17 April Serving files Scaling is harder ApacheDisk ApacheDisk ApacheDisk

Web 2.0 Expo, 17 April Serving files This doesn’t scale well Primary storage is expensive –And takes a lot of space In many systems, we only access a small number of files most of the time

Web 2.0 Expo, 17 April Caching Insert caches between the storage and serving nodes Cache frequently accessed content to reduce reads on the storage nodes Software (Squid, mod_cache) Hardware (Netcache, Cacheflow)

Web 2.0 Expo, 17 April Why it works Keep a smaller working set Use faster hardware –Lots of RAM –SCSI –Outer edge of disks (ZCAV) Use more duplicates –Cheaper, since they’re smaller

Web 2.0 Expo, 17 April Two models Layer 4 –‘Simple’ balanced cache –Objects in multiple caches –Good for few objects requested many times Layer 7 –URL balances cache –Objects in a single cache –Good for many objects requested a few times

Web 2.0 Expo, 17 April Replacement policies LRU – Least recently used GDSF – Greedy dual size frequency LFUDA – Least frequently used with dynamic aging All have advantages and disadvantages Performance varies greatly with each

Web 2.0 Expo, 17 April Cache Churn How long do objects typically stay in cache? If it gets too short, we’re doing badly –But it depends on your traffic profile Make the cached object store larger

Web 2.0 Expo, 17 April Problems Caching has some problems: –Invalidation is hard –Replacement is dumb (even LFUDA) Avoiding caching makes your life (somewhat) easier

Web 2.0 Expo, 17 April CDN – Content Delivery Network Akamai, Savvis, Mirror Image Internet, etc Caches operated by other people –Already in-place –In lots of places GSLB/DNS balancing

Web 2.0 Expo, 17 April Edge networks Origin

Web 2.0 Expo, 17 April Edge networks Origin Cache

Web 2.0 Expo, 17 April CDN Models Simple model –You push content to them, they serve it Reverse proxy model –You publish content on an origin, they proxy and cache it

Web 2.0 Expo, 17 April CDN Invalidation You don’t control the caches –Just like those awful ISP ones Once something is cached by a CDN, assume it can never change –Nothing can be deleted –Nothing can be modified

Web 2.0 Expo, 17 April Versioning When you start to cache things, you need to care about versioning –Invalidation & Expiry –Naming & Sync

Web 2.0 Expo, 17 April Cache Invalidation If you control the caches, invalidation is possible But remember ISP and client caches Remove deleted content explicitly –Avoid users finding old content –Save cache space

Web 2.0 Expo, 17 April Cache versioning Simple rule of thumb: –If an item is modified, change its name (URL) This can be independent of the file system!

Web 2.0 Expo, 17 April Virtual versioning Database indicates version 3 of file Web app writes version number into URL Request comes through cache and is cached with the versioned URL mod_rewrite converts versioned URL to path Version 3 example.com/foo_3.jpg Cached: foo_3.jpg foo_3.jpg -> foo.jpg

Web 2.0 Expo, 17 April Authentication Authentication inline layer –Apache / perlbal Authentication sideline –ICP (CARP/HTCP) Authentication by URL –FlickrFS

Web 2.0 Expo, 17 April Auth layer Authenticator sits between client and storage Typically built into the cache software Cache Authenticator Origin

Web 2.0 Expo, 17 April Auth sideline Authenticator sits beside the cache Lightweight protocol used for authenticator Cache Authenticator Origin

Web 2.0 Expo, 17 April Auth by URL Someone else performs authentication and gives URLs to client (typically the web app) URLs hold the ‘keys’ for accessing files CacheOriginWeb Server

Web 2.0 Expo, 17 April BCP

Web 2.0 Expo, 17 April Business Continuity Planning How can I deal with the unexpected? –The core of BCP Redundancy Replication

Web 2.0 Expo, 17 April Reality On a long enough timescale, anything that can fail, will fail Of course, everything can fail True reliability comes only through redundancy

Web 2.0 Expo, 17 April Reality Define your own SLAs How long can you afford to be down? How manual is the recovery process? How far can you roll back? How many $node boxes can fail at once?

Web 2.0 Expo, 17 April Failure scenarios Disk failure Storage array failure Storage head failure Fabric failure Metadata node failure Power outage Routing outage

Web 2.0 Expo, 17 April Reliable by design RAID avoids disk failures, but not head or fabric failures Duplicated nodes avoid host and fabric failures, but not routing or power failures Dual-colo avoids routing and power failures, but may need duplication too

Web 2.0 Expo, 17 April Tend to all points in the stack Going dual-colo: great Taking a whole colo offline because of a single failed disk: bad We need a combination of these

Web 2.0 Expo, 17 April Recovery times BCP is not just about continuing when things fail How can we restore after they come back? Host and colo level syncing –replication queuing Host and colo level rebuilding

Web 2.0 Expo, 17 April Reliable Reads & Writes Reliable reads are easy –2 or more copies of files Reliable writes are harder –Write 2 copies at once –But what do we do when we can’t write to one?

Web 2.0 Expo, 17 April Dual writes Queue up data to be written –Where? –Needs itself to be reliable Queue up journal of changes –And then read data from the disk whose write succeeded Duplicate whole volume after failure –Slow!

Web 2.0 Expo, 17 April Cost

Web 2.0 Expo, 17 April Judging cost Per GB? Per GB upfront and per year Not as simple as you’d hope –How about an example

Web 2.0 Expo, 17 April Hardware costs Cost of hardware Usable GB Single Cost

Web 2.0 Expo, 17 April Power costs Cost of power per year Usable GB Recurring Cost

Web 2.0 Expo, 17 April Power costs Power installation cost Usable GB Single Cost

Web 2.0 Expo, 17 April Space costs Cost per U Usable GB [ ] U’s needed (inc network) x Recurring Cost

Web 2.0 Expo, 17 April Network costs Cost of network gear Usable GB Single Cost

Web 2.0 Expo, 17 April Misc costs Support contracts + spare disks Usable GB + bus adaptors + cables [ ] Single & Recurring Costs

Web 2.0 Expo, 17 April Human costs Admin cost per node Node count x Recurring Cost Usable GB [ ]

Web 2.0 Expo, 17 April TCO Total cost of ownership in two parts –Upfront –Ongoing Architecture plays a huge part in costing –Don’t get tied to hardware –Allow heterogeneity –Move with the market

(fin)

Web 2.0 Expo, 17 April Photo credits flickr.com/photos/ebright/ / flickr.com/photos/thomashawk/ / flickr.com/photos/tom-carden/ / flickr.com/photos/sillydog/ / flickr.com/photos/foreversouls/ / flickr.com/photos/julianb/324897/ flickr.com/photos/primejunta/ / flickr.com/photos/whatknot/ / flickr.com/photos/dcjohn/ /

Web 2.0 Expo, 17 April You can find these slides online: iamcal.com/talks/