It would be good to make some network measurements (e.g. RTT, jitter, loss etc.) before, during and afterwards to see if we can quantify the effects. There may also be some measurements that can be made of various services.
-----Original Message-----
From: [log in to unmask] [mailto:[log in to unmask]] On Behalf Of Yang, Wei
Sent: Sunday, December 30, 2007 9:39 PM
To: Stephen J. Gowdy
Cc: Dubois-Felsmann, Gregory; atlas-sccs-planning-l
Subject: RE: [Racf-storagemgmt-l] Throughput update
That is always doable.
--
Wei Yang | [log in to unmask] | 650-926-3338(O)
> -----Original Message-----
> From: Stephen J. Gowdy [mailto:[log in to unmask]]
> Sent: Saturday, December 29, 2007 12:48 AM
> To: Yang, Wei
> Cc: Dubois-Felsmann, Gregory; atlas-sccs-planning-l
> Subject: RE: [Racf-storagemgmt-l] Throughput update
>
> Perhaps we can establish if we can pull out the test quickly if
> problems arise?
>
> On Fri, 28 Dec 2007, Yang, Wei wrote:
>
> > I will certainly send notification out in advance of this
> test. Our goal is to achieve 200MB/sec between SLAC and BNL in the
> first step. So ideally we want to have a 10Gb network so that we can
> do more than 50MB/sec. But I do not know when will we have 10Gb. I
> think we still have a chance to have 10Gb setup before the test. But
> we can't wait for the upgrade without participating any of such tests.
> I hope this test, if we have to do it under 1Gb, will not have too
> much negative impact to the rest of the lab. And I hope this e-mail
> exchange will serve as a call for 10Gb network upgrade ASAP.
> >
> > regards,
> > --
> > Wei Yang | [log in to unmask] | 650-926-3338(O)
> >
> >> -----Original Message-----
> >> From: Gregory P. Dubois-Felsmann [mailto:[log in to unmask]]
> >> Sent: Friday, December 28, 2007 6:51 PM
> >> To: Halperin, John H.
> >> Cc: Yang, Wei; Cottrell, Les; Buhrmaster, Gary
> >> Subject: RE: [Racf-storagemgmt-l] Throughput update
> >>
> >> Certainly we must be notified before such a test, so that
> Wilko and
> >> others can properly diagnose any problems it might cause.
> >>
> >> I would very much prefer to wait for the 10GB upgrade, if possible.
> >>
> >> Gregory
> >>
> >> On Fri, 28 Dec 2007, John Halperin wrote:
> >>> It's not clear to me whether you're planning on doing this test
> >>> (50MB/s for 24hr) before the SLAC link is upgraded. If
> so, do you
> >>> know if such a test will impact other users, eg, Babar data
> >>> import/export?
> >>>
> >>> --
> >>> John Halperin <[log in to unmask]>
> >>> SLAC Computer Security Team (Stanford Linear Accelerator Center)
> >>>
> >>> ------------------ Prev Msgs ------------------
> >>> Date: Fri, 28 Dec 2007 14:54:57 -0800
> >>> From: "Yang, Wei" <[log in to unmask]>
> >>> To: "Ernst, Michael" <[log in to unmask]>, Hironori Ito
> >>> <[log in to unmask]>
> >>> Cc: Jay Packard <[log in to unmask]>,
> >>> "Katramatos, Dimitrios" <[log in to unmask]>,
> >>> Joe Urbanski <[log in to unmask]>,
> >>> Rob Gardner <[log in to unmask]>, Dantong Yu <[log in to unmask]>,
> >>> "McKee, Shawn" <[log in to unmask]>,
> >>> Charles G Waldman <[log in to unmask]>,
> >>> RACF-STORAGE <[log in to unmask]>
> >>> Subject: RE: [Racf-storagemgmt-l] Throughput update
> >>>
> >>> SLAC is very interested in participating this test. There
> is a few
> >>> hurdles right now:
> >>>
> >>> 1) SLAC's external network is still 1GB. This limited the
> >> bandwide we
> >>> can contribute. However, I still hope we can at least contribute
> >>> 50MByte/s. We are planning to upgrade to 10GB in January.
> >>>
> >>> 2) For 50MB/s, a single GridFTP server is enough. But I
> hope we can
> >>> use SRM frontend. I am still waiting for a FTS channel to
> be setup
> >>> between BNL and SLAC SRM v2.2. There is no technical
> >> difficulty here.
> >>> The BNL-STAR channel already works with a testing SRM v2.2
> >> deployed on
> >>> SLAC's gatekeeper (However, we shouldn't run a production
> SRM on a
> >>> production gatekeeper).
> >>>
> >>> 3) We are looking for additional machines as GridFTP
> >> servers. Before
> >>> that will happen, I have borrowed two machines to run
> GridFTP. When
> >>> the lab reopens, we can have them setup.
> >>>
> >>> I hope we can get all these work before the middle of January.
> >>>
> >>> --
> >>> Wei Yang | [log in to unmask] | 650-926-3338(O)
> >>>
> >>> -----Original Message-----
> >>> From: Ernst, Michael [mailto:[log in to unmask]]
> >>> Sent: Friday, December 28, 2007 1:29 PM
> >>> To: Hironori Ito
> >>> Cc: Jay Packard; Katramatos, Dimitrios; Joe Urbanski; Rob
> Gardner;
> >>> Dantong Yu; Yang, Wei; McKee, Shawn; Charles G Waldman;
> RACF-STORAGE
> >>> Subject: RE: [Racf-storagemgmt-l] Throughput update
> >>>
> >>> Thanks, Hiro.
> >>>
> >>> Yes, I kept watching the ganglia graphs while you were
> running the
> >>> tests and the results are quite encouraging.
> >>> Though we could argue the goal of having 3 sites is
> >> fulfilled ('guess
> >>> you did transfers to UM and MSU apart from UC?) I am
> still eager to
> >>> demonstrate we can do these transfers to 3 Tier-2s. Also,
> >> it would be
> >>> desirable to not only show the aggregate rate out of BNL
> >> but also the
> >>> rates at the receiving end. Last point is, we managed to
> >> achieve this
> >>> rate for only 20 minutes (or, you presumably stopped the
> test after
> >>> ~20 minutes). This is not convincing, I'm afraid, we need
> >> to show at
> >>> least 12 hours, 24 hours would even be better.
> >>>
> >>> Again, I appreciate the effort spent by everybody helping
> >> to achieve
> >>> this. Please take a look whether we can take a step further
> >> to raise
> >>> the level of confidence regarding our capabilities on our
> >> end, and to
> >>> please our funding agencies ...
> >>>
> >>> --
> >>> Michael
> >>> -----Original Message-----
> >>> From: Hironori Ito [mailto:[log in to unmask]]
> >>> Sent: Friday, December 28, 2007 4:14 PM
> >>> To: Ernst, Michael
> >>> Cc: Jay Packard; Katramatos, Dimitrios; Joe Urbanski; Rob
> Gardner;
> >>> Dantong Yu; Wei Yang; McKee, Shawn; Charles G Waldman;
> RACF-STORAGE
> >>> Subject: Re: [Racf-storagemgmt-l] Throughput update
> >>>
> >>> Hello.
> >>>
> >>> I wanted to get 600MB/s. But, since I don't get it, here is
> >> what I got
> >>> today. I got about 500MB/s by transferring to UM (two different
> >>> places) and UC. Although, it shows the entire transfer
> to/from BNL,
> >>> they are almost entirely from BNL to other sites since the dCache
> >>> write pools at BNL shows not much activity.
> >>>
> >>> Is this good enough?
> >>>
> >>> Hiro
> >>>
> >>> Ernst, Michael wrote:
> >>>>
> >>>> Yes, 500 MB/s combined at whatever distribution.
> >>>>
> >>>> --
> >>>>
> >>>> Michael
> >>>>
> >>>>
> >>> --------------------------------------------------------------
> >>>>
> >>>> *From:* Jay Packard [mailto:[log in to unmask]]
> >>>> *Sent:* Thursday, December 27, 2007 5:42 PM
> >>>> *To:* Ernst, Michael
> >>>> *Cc:* Hironori Ito; McKee, Shawn; Katramatos, Dimitrios;
> >>> Joe Urbanski;
> >>>> Rob Gardner; Dantong Yu; Charles G Waldman;
> RACF-STORAGE; Wei Yang
> >>>> *Subject:* Re: [Racf-storagemgmt-l] Throughput update
> >>>>
> >>>> Michael,
> >>>>
> >>>> Just to verify, you would like a total of 500 MB/s to
> >> these 3 sites
> >>>> combined (rather than 500 MB/s to each for a total of 1500 MB/s)?
> >>>>
> >>>> Jay
> >>>>
> >>>> Ernst, Michael wrote:
> >>>>
> >>>> This looks good, indeed.
> >>>>
> >>>> Now I need you to plan for a transfer exercise to show that
> >>> this can
> >>>> be sustained
> >>>>
> >>>> over an extended period of time (let's say for 24hours)
> >>>>
> >>>> to 3 Tier-2 sites at the same time (possible candidates:
> >>> AGLT2, MWT2,
> >>> WT2)
> >>>>
> >>>> at a level of 500 MB/s out of BNL to these sites (at whatever
> >>>> distribution)
> >>>>
> >>>> We should make an effort getting this done before the
> >>> DOE/NSF Review
> >>>> in early February. I know this may be difficult at MWT2 and
> >>> SLAC but
> >>>> it's worth the effort, because the agencies are really
> >>> concerned about
> >>>> this point.
> >>>>
> >>>> --
> >>>>
> >>>> Michael
> >>>>
> >>>> --------------------------------------------------------------
> >>>>
> >>>> *From:* Jay Packard [mailto:[log in to unmask]]
> >>>> *Sent:* Thursday, December 27, 2007 4:40 PM
> >>>> *To:* Hironori Ito
> >>>> *Cc:* McKee, Shawn; Katramatos, Dimitrios; Rob Gardner;
> >> Joe Urbanski;
> >>>> Rob Gardner; Dantong Yu; Ernst, Michael; Charles G Waldman;
> >>>> RACF-STORAGE
> >>>> *Subject:* Re: [Racf-storagemgmt-l] Throughput update
> >>>>
> >>>> Oops, I sent the wrong graph - this is the correct one.
> >>>>
> >>>> Hironori Ito wrote:
> >>>>
> >>>> Hello.
> >>>>
> >>>> Your cacti is not showing the values I am expecting.
> >>>> Look at the attached plot from ganglia about between 13:20
> >> and 13:40.
> >>>> You see about 200-250MB/s increase in the traffic. This is the
> >>>> correct value since I know how much I sent.
> >>>>
> >>>> 70 (files)* 3600 (MB per files) / (20 minutes * 60
> >> sec/minutes) = 210
> >>>> MB/sec
> >>>>
> >>>> By the way, what was the 1st target value (200 or 250?)
> >>>>
> >>>> Hiro
> >>>
> >>
> >>
> >> --------------------------------------------------------------
> >> ----------------
> >> Gregory P. Dubois-Felsmann
> >> [log in to unmask]
> >> Experimental Physicist & BaBar Computing Coordinator
> >> tel: +1-650-926-4207
> >> Stanford Linear Accelerator Center, BaBar group
> >> fax: +1-650-926-3882
> >>
> >
>
> --
> /------------------------------------+-------------------------\
> |Stephen J. Gowdy, SLAC | CERN Office: 32-2-A22|
> |http://www.slac.stanford.edu/~gowdy/ | CH-1211 Geneva 23 |
> | | Switzerland |
> |EMail: [log in to unmask] | Tel: +41 22 767 5840 |
> \------------------------------------+-------------------------/
>
