TCLUG Archive
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
RE: [TCLUG:6469] Ebay downtime
actualy, it's not unheardof for the 10k to have a single domain board crash..
the first generation systems has turned up a large batch of defective CPU
modules (i remember 2 such bad modules in a 28 processor system i worked on)
the 10k was also designed by cray. similar systemboard interconnects
it could also be a DBA problem not a OS problem.. having to bring down an
oracle database. box is still there.. but the data isn't
On 14-Jun-99 lamfada@lugh.net wrote:
> On Mon, 14 Jun 1999 06:25:59 -0500 (CDT), Ben Kochie <ben@nerp.net> wrote:
>>I belive the MS case was specificaly against the Ultra Enterprise 10000
>>system.
>>
>
> If so, eBay needs better admins - these things are designed to be practically
> impossible to bring down.
>
>>if you know anything about the 10000, MS should stop whining and build a box
>>that can do as much as the 10k.
>>
>>64 processors?
>>hot swapable: CPU, RAM, system boards, I/O boards?
>>12gig/sec CPU->memory bandwidth?
>>
>
> screw the 10K, I want the SGI monster machine, Origin2000. Each machine is
> one
> to some number of nodes, basically up to 128 processor. It uses
> Cray-designed
> interconnects, which scale up as you put more CPU / disk / network into them.
> > admirably suited. But businesses and most consumers do not buy AMD because
> they have emotional investment in Intel. I admittedly run Intel chips, but I
> run some FP-math intensive processes, and would rather fork the cash for
> something with real power, like DEC Alpha or UltraSparc or even G3/G4 based.
>
> 128 processors
> 256GB memory
> 80GB/sec _sustained_ I/O bandwidth
> ccNUMA memory architecture
> Kick ass Iris as your console :)
>
> The 10K, OTOH, can be divided up into "virtual machines", up to 8, each of
> which
> gives the appearance of being a full standalone server. I do not know what
> exactly eBay's problem is, but I doubt it is the Sun box or even Slowaris,
> but
> probably their proprietary software or Oracle. Still, it is not strange that
> a
> company that can not build an OS that is stable or a database that doesn't
> corrupt itself would hold up the apparant failure of implementation of their
> competitors' solutions as proof that there must be something wrong with the
> products themselves. Their PR people are good enough to know that it really
> does not matter what the problem turns out to be, as long as they take their
> potshots now, the PHBs will hear those and ask the question "Why aren't we
> using NT?".
>
>>come on.. NT can't even use more than 16 CPU's
>>
>
> Yes, but put that (and all of the other arguments) in word the PHBs can
> understand, and are not bored by, and at the same time have emotional appeal
> -
> after all, few arguments are won by logic, and few products are sold by
> logic.
> Take the Pentium* vs K* vs WinChip vs Cyrix. Intel's only real advantage
> over
> these other chips is floating point performance. In many ways, the AMD chips
> are better, but in this arena, Intel is king. Yet thousands of people must
> have chips that say Intel on them for no logical reason. They don't do
> scientific / numerical processing. They do word processing. The only thing
> that most people do that benefits from fast floating point math is first
> person games, eg., Quake*, Half-life, etc., and for these purposes, the K6*
> is
> admirably suited. But businesses and most consumers do not buy AMD because
> they have emotional investment in Intel. I admittedly run Intel chips, but I
> run some FP-math intensive processes, and would rather fork the cash for
> something with real power, like DEC Alpha or UltraSparc or even G3/G4 based.
>
>
> -Chris
>
>>On 13-Jun-99 Bob Tanner wrote:
>>> Not really linux related, but unix related.
>>>
>>> Anyone know why Ebay went down? According to MS's website it is a problem
>>> with
>>> ebay's Sun Enterprise Servers and Solaris.
>>>
>>> Anyone confirm this?
>>>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: tclug-list-unsubscribe@listserv.real-time.com
> For additional commands, e-mail: tclug-list-help@listserv.real-time.com
> Try our website: http://tclug.real-time.com
----------------------------------
E-Mail: Ben Kochie <ben@nerp.net>
Date: 14-Jun-99
Time: 12:20:57
This message was sent by XFMail
----------------------------------