210 likes | 360 Views
Storage Update Entirely About Storage & SC4. GridPP 15 j.jensen@rl.ac.uk. Deployment. Preparing for SC4 All sites must have SRM No classics Aiming for SRM 2.1 Deployed in parallel with 1.1 DPM already provides a 2.1 interface. Status. (Nearly) all sites have SRM 1.1
E N D
Storage UpdateEntirely About Storage & SC4 GridPP 15 j.jensen@rl.ac.uk Storage Middleware
Deployment • Preparing for SC4 • All sites must have SRM • No classics • Aiming for SRM 2.1 • Deployed in parallel with 1.1 • DPM already provides a 2.1 interface Storage Middleware
Status • (Nearly) all sites have SRM 1.1 • Either DPM or dCache (or both) • Nearly? • Occasional staff or hardware problems • Those who don’t have one in prod’n have started installations and testing • All classics accounted for • Decommissioning or migrating (DPM) Storage Middleware
Status • Mailing list + community works well • ~45 members • Focus: SEs, and FTS/GridFTP/etc only when SE related • Lots of good stuff in the Wiki • Helped NGS deploy SE (DPM) • For gLite evaluation Storage Middleware
SRM 2.1 • Required for SC4 • Developed extensive test clients • Testing DPM’s 2.1 interface • Depend on FNAL delivering 2.1 for dCache • Says Timur: “mid Jan” Storage Middleware
dCache status • Upgrade • Most dCache Tier 2s have 1.5.{2,3} • Tier 1 had 1.6.5, now 1.6.6.1 • YAIM support • Wholly managed by GridPP • Good relations to DESY • 1.6.6.2 to go into LCG “2.7.0” • Upgrade Tier 2s to 1.6.6 • 1.6.6 has much improved logging Storage Middleware
DPM status • Good collaborations between GridPP and LCG • GIP plugin in LCG “2.7.0” • We’re testing the 2.1 interfaces • Upgrade Classic SEs to DPM experiences Storage Middleware
interoblems • SRM 2.1 is really SRM 2.1.1 • Two versions: 2.1.1 and 2.1.1-modified • 2.1.1 has a more soapy array handling • Most are now implementing 2.1.1-modified • DPM has switched, as of 1.4.1 • Publishing? • Both 1.1 and 2.1 are called “SRM” ? ? Storage Middleware
interoblems • Much of the API is optional • Some things are genuinely optional • Clients don’t use full API so why implement all • “Extra” features • Access control (StoRM (INFN)) • Interop is more than using same WSDL • API is quite complex (much more than 1.1) • Semantics not spec’d in WSDL Storage Middleware
interoblems SRM2 SRM2 put cycle srmCopy The Chocolate and Vanilla and Strawberry problem (cf. current dCache to DPM problems) Client srmCopy SRM2 SRM2 Storage Middleware
interoblems • Not all sites will upgrade at the same time • Some sites will continue to publish 1.1 • Even DPM has two different daemons • One for 1.1 and another one for 2.1 • So they compete for 8443 • Must be able to upload via one interface and download via the other Storage Middleware
Short Term Workaround Colourful solution Each flavour can talk to any other flavour Storage Middleware
Short Term Workaround Common client can talk to all flavours Storage Middleware
Use get/put for file transfer for now… GridFTP 3rd party xfer SRM2 SRM2 srmGet srmPut Client Storage Middleware
Remember Don't Panic Storage Middleware
Patience • SRMs will interoperate • SRMs are converging • Needs testing though • They will use 2.1.1-modified • They will be able to use srmCopy between SRMs Pessimist Optimist Storage Middleware
SC4 risks • Deployment/functionality/interoperability of SRM 2.1 delayed • Keep 1.1s around • Use FTS – supposed to be able to speak to all • dCache DPM transfer problem • Very serious – high impact • Problem not fully understood – developers informed • Problem with both lcg-* and FTS • Mitigation? Storage Middleware
SC4 risks • SRM2 for CASTOR2 (dev’d at RAL) • Will affect the Tier 1 – eventually (Julyish) • Critically important for CERN • CASTOR2/SRM2 + robot upgrade • Will affect Tier 1 – eventually • FTS [talking to SRM 2.1] delayed • Keep using SRM 1.1 • Migration from YAIM to gLite XML tool • Stick with YAIM for now… Storage Middleware
Metrics/Monitoring • Basic metrics: • # sites with SRMs deployed • Amount of storage published (c’ly 160 TB in T2) • Monitoring • Need for more extensive SE tests than SFT • Deploying SRM monitoring infrastructure • Exercising 1.1 up/download & 3rd party & del • Flexible – can eventually handle 1.1s and 2.1s Storage Middleware
Other Implementations • Not supported for Tier 2 deployment • CASTOR2 SRM • Developed at RAL • Deployed at CERN (other CASTOR sites later) • StoRM • Developed at INFN • Not tested in UK yet • Thin layer, depends on underlying filesystem Storage Middleware
Summary • All hands on SC4 preparations • Pretty well on track • Thanks to community working together • Some risks • Particularly external dependencies • Mitigating whenever possible • Taking pessimised view on interoblems • And don’t panic ! Storage Middleware