Attending: Brian, Chris B, Chris W, David, Duncan, Gareth, Jens, Jeremy, John B, John H, Elena, Lukasz, Matt D, Matt RB, Raja, Sam, Winnie

Special guest speakers: Martin Bly, Jonathan Perkin

0. Operational issues: xroot problem at Bristol - might it be the same as the xroot problems elsewhere (L'pool and RAL), or would it be different. Yesterday in dteam, the firewall was suggested. Also, as Chris discovered, the machine is on a 1Gb link which may well be saturated.

1. Martin Bly gave a report from HEPiX

· Site storage reports

· EOS being used (by CERN) across long distances, 200 M files, 99.4% availability, 22 PB across five instances.

· Addressing data corruption: two bits had flipped in a way which preserved the Adler sum. Adler was never meant to be a secure checksum. Do you need more checksums - or securer checksums - or maybe error correction, as Sam has experimented with at Glasgow (which would have the advantage of being able to correct bitflips)

· CERNBOX, a dropbox style, er, box cloud thingy - based on OwnCloud (not OwenCloud as I was about to type) - also Liverpool and RAL have OwnCloud instances.

· ZFS collaboration - OpenZFS now for Linux.

· SSD benchmarking at CERN

· BNL also using CEPH: indeed there seems to be enough interest in CEPH that it could become a topic in its own right at the next HEPiX.

· Clustered Netapp storage using NFS

· Chris asked about a Florida site using Kerberos to allow access but didn't have details in their presentation - maybe worth following up.

2. Jonathan had problems hearing us initially and then speaking: so we will have to postpone the pleasure of hearing from T2K till next week.

3. AOB

NOB, except it would seem the next HEPiX being in Oxford - and there being interest in CEPH - would be an opportunity for GridPP.

Jens Jensen: (22/10/2014 10:12:43)

Jon may be having vidyo trouble... he can't hear us.

Christopher J. Walker: (10:12 AM)

Someone is noisily typing

Jens Jensen: (10:13 AM)

sorry that's me. I've muted

Samuel Cadellin Skipsey: (10:17 AM)

Well, if you have enough files... (and ADLER32 is not a guaranteed secure checksum anyway)

Jens Jensen: (10:17 AM)

If cksum is wrong then something is wrong...

if _one_ cksum is wrong..

Samuel Cadellin Skipsey: (10:18 AM)

but you don't know *where* it's wrong, Jens.

This is *precisely* what erasure-coding schemes are designed to cope with.

(Did I mention that I have a beta plugin for EC in Dirac ;) )

Jens Jensen: (10:18 AM)

Cernbox is based on a cloud system which we're also running at RAL - name escapes me at the moment

Samuel Cadellin Skipsey: (10:19 AM)

(and, to be fair, xrootd does have an ec plugin nowadays too)

John Bland: (10:19 AM)

owncloud

we have one too. Can't get anyone to use the damn thing

Jens Jensen: (10:19 AM)

Yes, owncloud - thats it

NERSC also run iRODS IIRC - big instance

Sounds like an opportunity for GridPP: HEPiX in Oxford

Christopher J. Walker: (10:26 AM)

IIRC Adam Huffmann knows some of the CEPH developers - so there may be some way of inviting them.

Samuel Cadellin Skipsey: (10:26 AM)

Also, some other of the RAL guys are going to a Ceph thing today, so they will also know the Ceph devs then :D

Ewan Mac Mahon: (10:30 AM)

I think that counts as unusable,

I'm afraid.