EMC S210 Bedienungsanleitung

White Paper Abstract Most file systems ar e a thin layer of organizati on on top of a block device and cann ot effici ently address data at l arge scale. This paper focuses on OneFS, a modern fi le system that me ets the uniq ue needs of Big Data. OneFS i ncludes SmartPool s, a native tierin g capability, whi ch enables enterprises to reduce storag[...]

2 Next Generati on Storage Tiering wi th EMC Isil on SmartPools Copyright © 2015 EMC Corporati on. All Ri ghts Reserved. EMC beli eves the informati on in this publi cation is accurate as of its publi cation date. The informati on is subject to chang e without notice. The information i n this publi cation is provided “as i s .” EMC Corporation[...]

3 Next Generati on Storage Tiering wi th EMC Isil on SmartPools Table of Contents Introduction ................................................................................................... 5 Isilon Scale-Out NAS Architect ure .................................................................. 5 Singl e File System .............................[...]

4 Next Generati on Storage Tiering wi th EMC Isil on SmartPools Example D: Metadata P erformance for Sei smic Interpretati on ........................... 38 Conclusion ................................................................................................ .... 39 About EMC Isilon ............................................................[...]

5 Next Generati on Storage Tiering wi th EMC Isil on SmartPools Introduction EMC Isil on SmartPools enabl es multiple levels of performanc e, protectio n, and storage density to co-exi st wit hin the same file system, and unlocks the abi lity to aggregate and consolidat e a wide range of appli cations wi thin a single extensibl e, ubiquitous storag[...]

6 Next Generati on Storage Tiering wi th EMC Isil on SmartPools not only the ope rating system, but al so the underl ying file system that stores and manages data in the Isilon storage cl uster. Single Fil e System OneFS provi des a single fil e system that operates out side the constrai nts of traditi onal scale up construct s li ke RAID groups, a[...]

7 Next Generati on Storage Tiering wi th EMC Isil on SmartPools levels are dynami cally confi gurable at a per-fi le or per-fil e system granu larity, or anything in between. Data an d metadata acc ess and lockin g are coherent ac ross the cluster, and thi s symmetry i s fundamental to the simpli city and resili ency of OneFS’ shared nothi ng arc[...]

8 Next Generati on Storage Tiering wi th EMC Isil on SmartPools The process of cho osing a new l ayout is call ed restripi ng, and thi s mechanism is identical for repai r, rebal ance, and tiering . Data i s moved in th e background with the file availabl e at all times, a process that ’s completel y transparent to end users and applicati ons. Ca[...]

9 Next Generati on Storage Tiering wi th EMC Isil on SmartPools Figure 2: OneFS Caching Hierarchy Smart Pools Overview SmartPools i s a next-generati on data tierin g product that b uilds di rectly on the core OneFS attributes d escribed above . These i nclude the si ngle file system, extensib le data layout an d protection framew ork, parall el jo[...]

10 Next Generati on Storage Tiering wi th EMC Isil on SmartPools Figure 3. SmartPoo ls Tiering Model The current SmartPo ols impl ementation expands on th ese two core concepts in several additi onal ways, whil e maintaining simplicity and ease of use as primary goals. Storage Poo ls Storage Pools provi de the abili ty to define subsets of hardwar [...]

11 Next Generati on Storage Tiering wi th EMC Isil on SmartPools protection stri pes or mirrors don’t extend across di sk pools (the excepti on being a Global Namespac e Acceleration extra mi rror, described b elow). Disk pool s are managed by OneFS an d are not us er config urable. Figure 4. Cluste r Disk Pool Allocation Node Pool s Node Pools a[...]

12 Next Generati on Storage Tiering wi th EMC Isil on SmartPools Each Node Pool onl y contains di sk pools from the same typ e of storage nodes and a disk pool may bel ong to exactl y one node pool. For exampl e, S Series node s with 300 GB SAS drives and one 4 00 GB SSD per node w ould be in one node p ool, wh ereas NL Series wit h 4 TB SATA Drive[...]

13 Next Generati on Storage Tiering wi th EMC Isil on SmartPools Figure 5. Automatic Provisioning into Node Poo ls Tiers Tiers are groups of n ode pools combi ned in to a logical superset to opti mize data storage, accordi ng to OneFS pl atform type. This is ill ustrated in figure 6, below. For example, X Series node pools are oft en combined i nto[...]

14 Next Generati on Storage Tiering wi th EMC Isil on SmartPools Figure 6. SmartPools Tier Configurat ion SmartPools users typicall y deploy 2 to 4 tiers , with th e fastest tier usuall y contai ning SSD nodes for the m ost performanc e demandi ng portions of a work flow, and the lowest, capacit y-biased ti er comprisin g multi-TB SATA dri ve nodes[...]

15 Next Generati on Storage Tiering wi th EMC Isil on SmartPools directed toward s a subset of clu ster resources . Thi s is often referred to as “nod e bullying” and can man ifest as sev ere over-ut ilization of resources l ike CPU, network or disks. However, admin istrators can safely target specifi c data to broader class es of storage by cr[...]

16 Next Generati on Storage Tiering wi th EMC Isil on SmartPools Isilon Hundred -series Node Equivalent IQ-s eries Node X200 with 6GB RAM and 1 2 500GB HDDs IQ 6000x X200 with 6GB RAM and 12 1TB HDDs IQ 12000x X400 with 24GB RAM and 36 1TB HDDs IQ 36000x X400 with 24GB RAM and 36 2TB HDDs IQ 72000x X400 with 24GB RAM , 32 1TB HDDs & 4 100GB SSD[...]

17 Next Generati on Storage Tiering wi th EMC Isil on SmartPools drive types, quanti ties and capaci ties and c ompati ble RAM config urations to be provisioned i nto the same pools. SSD Compa tibility and Equivalence OneFS 7.2 and beyond also contain an SSD compatibility option, which allows nodes with dissimilar capacity SSDs to be provisioned to[...]

18 Next Generati on Storage Tiering wi th EMC Isil on SmartPools Supported prot ocols Data Sp ill Over If a Node Pool fill s up, writes to th at pool will automatically spill over to the next pool. This default behavior ensures that w ork can conti nue even if one type o f capacity is full. There are some circumstances in which spill over is undesi[...]

19 Next Generati on Storage Tiering wi th EMC Isil on SmartPools Figure 8. Autom atic Provi sioning with Sma rtPools When initi ally configuring a cl uster, OneFS aut omati cally assigns nodes to nod e pools. Nodes are not pool ed — not ass ociated with each other — un til at least th ree nodes are assigned to the n ode pool . A node pool wi th[...]

20 Next Generati on Storage Tiering wi th EMC Isil on SmartPools Figure 9. Glob al Name Space Acceleration The purpose of GNA is to accelerate th e performance of metadata -int ensive applicati ons and workloads such as hom e directori es, workflows wi th h eavy enumeration and acti vities requi ring a large numb er of comparisons. Examples of meta[...]

21 Next Generati on Storage Tiering wi th EMC Isil on SmartPools without adding additional configuration set ting s, the alternative is reservi ng an unreasonabl e percentage of the SSD spac e in a node pool . The default for new clusters is for Vi rtual H ot Spare to hav e both "reduc e amount of available space" and "d eny new data[...]

22 Next Generati on Storage Tiering wi th EMC Isil on SmartPools anything wri tten to path /i fs/foo goes to th e S Series node s in Node Pool 1, then moves to the NL Se ries nodes in Node Pool 3 when older than 30 days. To simplify manag ement, there are defau lts in pl ace for Node Pool and Fil e Pool settings whi ch handle basic d ata placement,[...]

23 Next Generati on Storage Tiering wi th EMC Isil on SmartPools Figure 10 . Sm artPools File Poo l Policy Eng ine File Poo l Policies SmartPools fi le pool policies can be us ed to broadl y control the three pri ncipal attribut es of a file. Namely: 1. Where a fil e resides. 1. Tier 2. Node Pool 2. The file performanc e profil e (I/O optimization [...]

24 Next Generati on Storage Tiering wi th EMC Isil on SmartPools A file pool poli cy is built on a fil e attribut e the policy can match on. Th e attribut es a file Pool poli cy can use are any of: Fil e Name, Path, Fil e Type, File Si ze, Modified Time, Create Time, Metadata Chan ge Time, Access Ti me or User Attri butes. Once the fil e attribute [...]

25 Next Generati on Storage Tiering wi th EMC Isil on SmartPools In practice, defaul t Fil e Pool policies are al most al ways used because th ey can be very powerful. Most admi ni strators do not want to set rul es to govern all their data. They are generall y concerned about some or most of t heir data i n terms of where it si ts and how accessi [...]

26 Next Generati on Storage Tiering wi th EMC Isil on SmartPools Figure 11 . Sm artPools File Poo l Policy govern ing data moveme nt Figure 12 . File Pool Policy Exam ple Figures 11 and 12 i llustrate a common Fil e Pools use case: An organization and want s the active data on th eir performanc e nodes in Tier 1 (SA S + SSD) and to move any data no[...]

27 Next Generati on Storage Tiering wi th EMC Isil on SmartPools File Pool polici es can be created, copi ed, modified, priori tized or removed at any time. Sample poli cies are also provi ded that can be used as i s or as templates for customization. Custom File A ttributes Custom Fil e Attributes, or user attrib utes, can be used wh en more granu[...]

28 Next Generati on Storage Tiering wi th EMC Isil on SmartPools The engin e can also be run on -demand usi ng a separate invocati on to a ppl y the appropriate fil e-pool membershi p settin gs to an individual fil e or subdirectory with out having to wait for the background scan t o do it. To test how a new p oli cy will affect file disposition s,[...]

29 Next Generati on Storage Tiering wi th EMC Isil on SmartPools that file matches a fi le pool poli cy which is based on any other criteria besi des path name, SmartPool s will write tha t fil e to the Node Pool wi th the most avail able capacity. If the fi le matches a fil e pool policy that places it on a di fferent Node Po ol than the hi ghest [...]

30 Next Generati on Storage Tiering wi th EMC Isil on SmartPools Using Sma rtPools to Improve Perform ance SmartPools can b e used to imp rove performance in several ways:  Location-based Perform ance  Performance Setti ngs  SSD Strategies  Performance Isolat ion Location-based performance leverag es SmartPool s file pool poli cies to c[...]

31 Next Generati on Storage Tiering wi th EMC Isil on SmartPools Data Acce ss Settings At the File Pool (or even the singl e file) level, Data Access Setti ngs can be confi gured to optimi ze data access for the type of app lication accessin g it. Data can be optimized for Concu rrent, Streami ng or Random acces s. Each one of th ese settings chang[...]

32 Next Generati on Storage Tiering wi th EMC Isil on SmartPools  Metadata read acc elerati on: Creates a preferred mirror of fi le metadata on SSD and writes the rest of the m etadata, plu s all the actual file data, to H DDs.  Metadata read & w rite accelerati on: Creates all the mirrors of a fil e’s metadata on SSD.  Avoid SSDs: N[...]

33 Next Generati on Storage Tiering wi th EMC Isil on SmartPools The SmartPools Setti ngs page of the WebU I al so provides th e ability to wh ether L3 cache is the defau lt configu ration for any new nodepools th at contai n SSD. Figure 16. Co nfiguring L3 C aching by defa ult on new Nod epools. When convertin g the SSDs in a parti cular nodepool [...]

34 Next Generati on Storage Tiering wi th EMC Isil on SmartPools balanced to, anoth er class of n ode. This access n ode may have di fferent front end IO capabiliti es, cache, etc, but th e performance characteristi cs will still primarily be governed by the cl ass of storage on which the data resi des. A good rule of thumb is that whil e SAS is go[...]

35 Next Generati on Storage Tiering wi th EMC Isil on SmartPools  Ensure that SSDs c omprise a mi nimum of 2% o f the total cl uster u sable capacit y , spread across at l east 20% of the nod es, befo re enabling GNA.  Determine if metad ata operati ons for a particul ar workload are bi ased towards reads, writes, or an even mi x, and select [...]

36 Next Generati on Storage Tiering wi th EMC Isil on SmartPools Post producti on creates a massive a mount of dat a as their main product . This can grow to 100s of terabytes and often mu ltiple p etabytes, and needs t o be accessed very quickl y by many people and proc esses whil e a project is active. Th en it continues i ts rich commercial life[...]

37 Next Generati on Storage Tiering wi th EMC Isil on SmartPools Figure 18 . D ata Performance & Protection i n EDA These engin eering design data and hi storical design data examples are val uable because they i llustrate the need for dat a protecti on for two sets of data that exist si de by side in th e same infrastructure, one o f which i s[...]

38 Next Generati on Storage Tiering wi th EMC Isil on SmartPools Because this i s a demandi ng production environ ment where data l ife is very long, there is a strong p reference for scalabl e, durable systems that combine performan ce and archive capabi lities wit h a compelling cost of own ership. In this case, the pref erred archit ecture would[...]

39 Next Generati on Storage Tiering wi th EMC Isil on SmartPools All of the metadata i s stored on SSD, but dat a may be relegated to faster or slow er disk based on i ts relative timeli ness. The result is the abili ty to instan tly locate any data, while still being abl e to cost reduce the majori ty of the data on t he system. Because metadata i[...]